What's New

October 14, 2026

The cookie consent banner no longer reappears on every page for users whose saved preferences came from an earlier version of the consent settings. The app clears outdated preferences on a later page load, so the next accept is saved correctly.

October 13, 2026

Account & Onboarding

Four subscription tiers are now available: Creator, Pro, Studio, and Enterprise. Creator is the default for any organization without a subscription, Pro supports up to 100 seats, and Enterprise has no seat cap. Credits renew once per billing period — or once per calendar month for an organization without a subscription — at the exact minute the period ends. The usage panel shows the renewal date, time, and time zone.

Credits are now shared across all of an organization's brands: every brand sees the same balance, and spending in any brand reduces it for the rest. The credit unit is now minutes of video — uploading or importing a minute of video costs one credit, and generating a clip costs one credit. Cue chat, covers, reframes, and other actions no longer count against credits.

Credits used past the plan's allowance are now billed. Each action over the limit is reported to the billing system within about a minute, so the organization's invoice reflects actual usage.

The credits badge, the account usage panel, and the message shown when an action would exceed the remaining balance now include the overage cap the organization's owner has set.

Credit estimates shown before an upload or import now use the same per-minute rate as the actual charge.

Calendar

The Calendar now shows days in your own time zone instead of Pacific time, so "today", each post's day, and the weekly report ribbons match the day you see.

The Calendar header now has a Weekly report switch. When the report is off, hovering the switch shows what it is and offers to turn it on without leaving the Calendar; when it's on, turning the switch off pauses it.

Weekly report ribbons are back on the Calendar. A day a report lands on shows a corner ribbon — solid green once the report is stored, tinted while it is still to come — and selecting that day shows the report under the list. A stored report opens in a dialog with a download link; a scheduled one opens its settings in the Automate tab.

Automations

The weekly report now completes more reliably for brands with many posts. It reads all the data it needs in one pass, cutting the number of steps a busy week requires, and runs for up to 13 minutes instead of 5.

New weekly reports are spread across the weekend. A brand that turns the report on without choosing a time gets its own hourly slot on Saturday or Sunday, 7 AM to 9 PM in the creator's time zone, instead of every new report running at the same hour.

The weekly report no longer runs for a week when none of the brand's own videos went out. The first report arrives after the brand's first published video.

The weekly report no longer recommends or analyzes the sample clips every new brand starts with.

Cue

Cue chat is now free on every plan. The credit gate and the banner that said the brand had used all its credits are gone.

Your messages in a Cue conversation now appear on the right in blue, while Cue's replies stay on the left. Images you attach line up under your message on the right.

Cue's chat bubbles sit closer together, and Cue's replies use a darker gray bubble so the bubble stands out from the page behind it.

Fixed the handoff from a clip plan to its build: windows that share a group now build as one stitched clip, the plan card shows them as parts of one clip and counts clips rather than windows, and the card no longer fails when the model sends the window list as a JSON string.

Discover

The Discover selection bar now puts Create compilation and Generate short-form clips behind one Create button whose menu opens on hover or press. Find ideas in this video is now a Scan for more button, and board saves are one Save to Board button with a menu that lists the brand's boards.

Discover's Cue column is narrower, giving the results more room. An empty Cue thread now opens with a message from Cue asking what to look for, in place of a centered help paragraph.

Each clip idea tile in the Discover grid now has Clip it and Save directly under it. Clip it starts a clip from the idea's moment in one press; Save puts the idea on a board. An idea already made into a clip shows Save alone.

The Search button in Discover now hides while the search field is empty, appearing only once you type a query.

Discover's selection bar now sits 48 pixels above the bottom edge of the page.

Fixed Discover scrolling sideways in a narrower window. The results and the Cue column now shrink to fit.

Video Editor

The video editor header now has a Go to post link beside Save Updates, opening the post that uses this video in one step with a back-link to return to the editor. The link appears only once the edit is saved and only when the video already has a post.

The transcript's word toolbar now opens for a single selected word, so one hidden word can be shown again and one cut word can have its video restored without clearing the rest of its line. Fixed a word coming back blank or missing its capital letters and punctuation when shown after being hidden along with its whole line.

Fixed hook text, end text, and captions in a brand's uploaded font running past the edges of the video preview when the font arrived after the initial layout pass.

Clips & Content Library

Clip rows in the content table now show each clip's thumbnail at the clip's actual shape, so a landscape or square clip no longer reads as a tall crop of itself.

The cold open mark — a lightning bolt — now appears beside a clip's name in the Content list row, the same mark the details pane already shows.

Large video uploads now finish faster, sending more of the file at once in fewer and larger pieces.

Fixed the TikTok and Instagram import picker listing only a connection's first linked account. It now lists each account the connection links, so a video from any of them imports and links to its correct post.

Fixed imports and uploads of video files labeled with a generic binary content type. Video extensions now identify those files, and uploads keep the video content type the processing pipeline needs.

The TikTok and Instagram import picker now refuses a second import while one from the same platform is still running, with a toast that says so.

Creator Watchlist

A watched creator's detail page now has a Refresh now button, under the last and next refresh times, that works the same as the watchlist row menu's refresh with the same messages and pending state.

YouTube follower counts are rounded to 3 significant figures, so two equal rounded counts now read "No visible change" in the watchlist and on the creator page instead of "0". A change between two different rounded counts is marked approximate with "≈", and the creator page says why under the change.

Fixed Refresh now showing "Refresh again in 25 hours" right after a creator refresh completed. The wait shown is now never more than the 24-hour cooldown.

Fixed the Refresh now item in a watched creator's row menu reading as available during its cooldown. The caption now shows in full, including when Refresh now opens again, and the item no longer highlights under the mouse.

Playbook

Each Playbook card now leads with what Tempo uses for that slot and why. The card lists each active instruction with where it came from and what the brand's own posts say about it; defaults the brand's results overturned are folded under "Not used on your channel" with the verdict as the reason.

The Playbook tab no longer appears in a brand's sidebar by default.

Brands & Connections

Brands now have a Category setting in brand settings. Tempo infers the category from the brand's own videos, or from posts on connected platforms when no videos exist, and the Playbook says which source it used. A creator can pin the category instead, or hand it back to Tempo.

AI tools connected to Radical Tempo can now read and manage organizations: renaming an organization, reading its active brand count, and navigating from an organization to its brands. A brand's name, defaults, and style guide can also be updated through an AI tool, with protection against overwriting a change made in another session at the same time.

A team member invited to a single brand no longer sees the organization's full brand count — the count now reflects only the brands that member can access.

Docs, Site & Integrations

The documentation has a new look — brand fonts, a mint canvas with a dot grid, green accents, and rounded cards. The home page now opens with a short introduction and cards linking to Quick start, Connect an AI tool, Importing videos, Publishing, and the Tool reference.

A new Connect an AI tool page gives the server URL and setup steps for Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code, and Gemini CLI, with example prompts and an FAQ. A new Tool reference page lists every action a connected AI tool can take, grouped by job, generated from the server's own registrations so it stays current.

The Radical Tempo homepage now has a section on connecting from Claude, ChatGPT, Gemini CLI, Cursor, and VS Code: the server URL with a copy button, a link to the setup guide, and an example of a publish proposal waiting for approval.

A connected AI tool that needs to sign in now receives an OAuth challenge instead of a tool error, so Claude.ai and ChatGPT can start the sign-in flow on their own.

Generating clips from a connected AI tool now uses one generate_clips command, starting from a prompt, from exact moments, or from clip ideas, with a review flag that plans clips for your approval before cutting them.

AI tools that access Radical Tempo through tool calls can now list and read brands, videos, posts, chapters, transcripts, and memories through dedicated commands.

Agents can now set a post's hook text. update_post gains a hookText field that sets or removes the on-screen hook for one post, and a hookTextAllPlatforms field that applies that hook to every post of the video, matching the post editor's toggle. When a post is blocked from publishing, propose_publish now names the specific field behind each problem.

update_post also gains a caption field that writes the text the platform publishes first — the title on YouTube and Facebook videos, and the caption on every other post type. It refuses a title on a post type that publishes none, rejects all text on a Snapchat story, and rejects a description that would exceed the platform's character limit once hashtags are appended.

Agents can now import a Google Drive file shared with anyone who has the link. Pass a drive.google.com share URL to import_from_url and the file downloads directly; a file that is not shared this way is refused with a message that says so rather than a generic content-type error. Imported video titles now come from the filename the download sends, so a Google Drive link no longer arrives named "download".

Fixed sign-in from Gemini CLI, which failed because the protected-resource metadata did not match the URL it connected to.

Fixed sign-in from Gemini CLI after the user approved access, which failed with "Missing 'iss' parameter in authorization response per RFC 9207". The consent redirect now includes the issuer parameter.

Fixed sign-in from Codex, which failed with "OAuth authorization server issuer does not match authorization metadata origin". The protected-resource metadata now reports the authorization server by the issuer RFC 8414 clients require.

Fixes

Every page in a brand's sidebar — Search, Automate, Connect, Settings, and Playbook — now has the same title size, weight, and position as the Content page.

Floating toolbars now have a visible border and a deeper shadow so they stand out from the white content behind them. This covers the selection bars in Discover and the content library, and the transcript, scene, and trim toolbars in the video editor.

A retried automation job no longer charges a brand's credits twice for the same work.

Fixed the Approve button on Cue's publish, transcript deletion, automation, and compilation proposal cards when used through an AI tool host that renders cards without declaring MCP Apps support.

Plan and usage displays now refresh when an organization's plan changes, and pending invitees no longer see billing details.

The Performance pages now load up to 5 seconds faster for brands with large post histories.

The company logos on the Radical Tempo homepage are now centered on phones and tablets.

October 9, 2026

Fixed a video that kept playing after you navigated away from its page. Players on the post page, the video library, the video editor, the template editor, and search now pause when you leave.

October 6, 2026

Playbook

The Playbook page title now reads "Tempo's playbook for [brand name]". The back link, the checked date, and the Playbook eyebrow at the top of the page are gone.

Each Playbook card now has an expandable section that lists the slot's other instructions with their status and evidence directly on the card. The Why panel explains the headline instruction alone — its provenance, history, and the audit link.

Structure, Premise, and Opening cards now list the brand's own segments, people, topics, and phrases under "On your channel", with how each performed against the brand's other posts on the same platform, the number of posts behind the result, and a Correlation tag. Opening the Why panel for one of those rows adds the measured difference and its range, how many weekly checks the result has held, and how much of the measurement Tempo's own clips account for.

Discover

Every card in the Discover grid now has the same shape — a frame with its timecode, a play button and checkbox on hover, one row of badges, and a title with snippet — whether the result is a spoken moment, an on-screen moment, a clip, a chapter, or a clip idea. A long "matched" badge truncates instead of pushing Keep onto a second line. The bulk-selection toolbar no longer wraps a button label to one word per line.

Publishing

The "What shaped these suggestions" section in the post editor is now collapsed behind its heading. Opening the heading expands the list, the same way the Why this clip panel works on a clip.

Insights

Performance reports now name an untitled clip by the caption's headline — for example "Payback looks best in pink" — rather than repeating the full caption with its call to watch and hashtags. Story moments now start 3 seconds into a clip, so an early retention drop is not treated as the story's key moment.

Fixes

Hook text, end text, and captions no longer run past the edges of the video when the font had not yet loaded at the time of the first layout pass. Bold, italic, uppercase, outlined, and boxed captions that could come out wider than the video or wrap a word onto a second line are also fixed.

September 29, 2026

Clips & Content Library

The Generate Clips window for a video that has finished processing has a new layout. The header names the video; the left column holds the source video, then the video summary and director's notes as folding sections; and the right column holds the video's format and category, a Subjects tab, a Chapters tab, and the clip settings — "Use smart settings" or "Let me choose", the language picker, and a Generate Clips button spanning the bottom.

The Generate Clips window for a video still processing follows the same design: a header naming the video, the thumbnail spanning the full window width, director's notes that fold from a bar under the thumbnail, and the same smart-settings toggle with the framing, speed, and max duration pickers visible only when you choose to set them yourself.

Search

Search now finds moments from what happens visually in a video, not only what is spoken. The search index includes time-coded windows of each video's visual events alongside its transcript windows; a query like "slap" finds the shot where someone is slapped in silence, and the moment cues the player to that frame. Each result now says whether it came from the spoken words or from the visual track, and the "On screen" filter in Discover keeps a video whose visual window matched by meaning.

Search now matches other forms of a query word. A word of four or more characters finds any word that starts with it — "slap" finds "slaps", "slapped", and "slapping" in transcripts, visual events, titles, and chapters — with the matched word highlighted and the moment time-coded as for an exact match. Shorter words still match exactly, so "cat" does not match "category".

When a related word or phrase found a result in place of your query, a "matched" chip now appears on the result card, the grid tile, and the detail pane saying which one — for example "matched 'slap'" on a search for "smack" — so a highlighted word that differs from the query reads as part of the search rather than a mistake.

Search moments are more precise. A hit's best moment found by meaning is always kept, and further moments are kept only when they sit close to it, so the unrelated windows of a long video no longer crowd the list. A stretch of transcript that holds only a sound tag is no longer a moment, and a visual event over such a stretch now reports what is on screen instead.

Fixed Discover, the Search page, and Cue's search showing "Search is unavailable right now" for any search that matched a video's words directly.

Discover

The Discover grid tile now fills its full column rather than stopping at 160 pixels inside it, on both the search results grid and the chapters grid. Keep is now a small "Keep" chip right after the Exact or Topic badge — always visible below the frame — rather than a bookmark that appeared over the footage on hover.

The tile's hover preview is now a round play button named by the timecode, which appears as a tooltip, rather than a pill that covered most of the frame. Opening the preview shows the video's name with the match badge beside it, a timeline strip marking where in the video the moment sits, and a "Keep to board" button in the header.

Insights

The weekly performance report is now the Monday dispatch. It reads every video the brand published in the window — upload or clip — and produces a headline about what the platforms did, a trajectory block on which video was pushed and whether anything is still moving, one verdict card per video with the video's cover and platform breakdown, a line per funnel stage with the facts that read it and Cue's reading, a moment card with the words at the retention drop, and two or three moves in Radical Tempo's own terms.

The analyst curates the dispatch into one to three stories, each about one video or a named pair, written in the first person with the move each story leads to. Every video is still drawn in full — its stats, calendar, platform split, and a chart of its reach against its stay — and one line covers the videos no story picks up.

Every sentence the dispatch shows — each verdict stage, each trajectory item, each move, the headline, and the lede — is now written by the analyst with the figures in it. Each number is verified against the fact it came from; a sentence that does not pass is left out rather than stored. A part with no sentence left is omitted entirely, and the headline falls back to summarizing what remains.

Cue can now trace how each post's views arrived after it went out: its views by day, the share that came on its first day, and whether it is still moving. For YouTube posts, the breakdown also includes the non-subscriber share and the top countries. Cue now answers "did the platform push it" and "is it still growing" questions for any post, and the dispatch uses this data for every post in its window.

Captions

The video editor's caption settings now list one switch per transcript the video has, replacing the "Show a second caption set" switch. Up to three can be on at once; drag a row to move its language up or down in the stack, and the last switch cannot be turned off. Each set shows its own transcript's words in the editor preview, the video preview, and the rendered video.

The brand template editor's caption settings now control the default a video starts from: how many languages it shows (1 to 3), whether the video's own transcript always takes the top slot, and whether the newest transcript leads.

Each speaker now shows the same color in every stacked language. The color ranking was previously computed per language, so a speaker could appear in a different color in each translation. Colors are now assigned once, ranked by talk time in the transcript the video reads.

Video Editor

The video editor's timeline now always uses the zoomable mode. The toggle that switched back to the legacy timeline is gone; every creator gets the zoom slider, pinch-to-zoom, the caption track, and gap rows with "Add caption" in the transcript. A creator who had previously switched to the legacy timeline returns to the zoomable one the next time they open the editor.

The timeline ruler now spaces its ticks to match the playback speed, so a slowed-down edit's ruler stays readable instead of crowding ticks together.

Fixed a video download that handed back the raw source and dropped the brand's default template. A video the editor showed with the template's captions, end text, overlays, and styling would download without them if no template had ever been explicitly applied. Such a video now renders before downloading, so the file matches what the editor showed. Downloading the original source still returns the unmodified upload.

Fixed hook text and end text running into the sides of a video when a long line had to shrink to fit. The fit now measures the text as the video draws it — including its weight, style, capitalization, padding, and outline — so a long line keeps the full side margin in the editor preview and in the rendered video.

Publishing

Suggested captions, titles, and tags on the post editor page now arrive with the page. They previously required a second request that could only start after the page's main query finished; on a slow connection that added several seconds to the wait.

A post that uses the brand's default layout template now re-renders before publishing when the template was updated since the post's last render. Editing or setting the default template no longer leaves a published post showing the template the editor had before the change.

Cue

Cue's analytics can now compare a post's figures against the brand's own posts of the same kind on the same platform, age-matched to account for when each was published. At 24 or 48 hours the result states the post's views beside the peers' median at that age; at 72 hours it states the percentile among them. The weekly performance report includes this comparison for every post in its window.

Cue's performance trait comparison now reports on two opening traits: which kinds of on-screen text a video's opening shows — such as a title card or a premise overlay — and how its hook is delivered, whether spoken, visual, overlaid on screen, or in captions. Each trait appears as a fact sentence in the comparison results.

Fixed Cue potentially clearing a post's title or description when it edited something else. Changing a post's visibility or cover could empty its caption, or fail for emptying a title the platform requires. Cue now drops untouched optional fields before the call reaches the server.

Fixed the Approve button on Cue's compilation card. Cue previously refused the button, so pressing it had no effect and the compilation never started. The update Cue posts when a compilation finishes now names it "Compilation build" rather than an anonymous background task.

When Cue cannot reach Radical Tempo or cannot read a task's status, it now promises the user an update only for a task started in the same conversation, and says it will check again for any other.

Fixed Cue reporting that a transcript translation had failed to start when the translation had actually been queued. A failed read of the queued job returned an internal error; Cue now answers with the queued task instead.

Automations

The Clips on upload automation's Shape and Longest clip settings now include an Optimize option, letting the clip generator choose shape and length the way the Generate Clips dialog does by default. New automations start on Optimize for both settings. The Shape chips are also renamed Split, Original, and Adaptive, and Cue can request optimized settings when it generates clips through an automation.

Fixed the Automations tab calling a completed Publish analysis run "1 video". A run that produced a performance report now reads "Your report is ready.", and one that produced several reads "# reports are ready."

An automation run now includes the model calls it makes beside its main turn — such as the dispatch's fact-check passes and Cue's guardrail calls — in the credits it charges, the way a chat turn already did.

Fixes

Clip Ideas is no longer shown on Discover. Discover opens on Search; the mode switch and the "Clip ideas from this video" link on each result are gone.

Fixed Generate Clips saving a video's summary as the creator's own when nobody had edited it. A clip run now saves the summary only when the creator changed it, so a video processed again later shows its new summary in Generate Clips, the video details, and the next clip run.

Fixed the video player failing to render for videos with captions.

Fixed Discover not clearing a scope filter when using a GPT response model. Clearing a duration bound, a season, or a video scope now works correctly on all models.

The clip preview now correctly counts the playback speed you set when deciding whether overlays apply, matching what the rendered video will show.

Fixed the "No Crop" framing option showing an empty icon for a video whose dimensions were not yet known. It now shows a widescreen placeholder.

Fixed the framing and speed pickers in Generate Clips losing keyboard focus when a pick moved them onto or off the recommended setting.

Fixed a video keeping its old transcript words after a save that changed them. When saving the video's other edits failed behind the scenes, the new words were discarded with them even though the save reported success. The new words are now kept when the rest of the save fails.

Fixed adding an image layer to a video that had never been edited. The layer now applies to the whole, untrimmed video.

September 22, 2026

Discover

Discover's page has a new layout. All controls share a single header: the search field with Find moments inside it, a Find ideas button beside the field, the look-back window for a run, and the brand's boards as chips. Results and Ideas take small uppercase section headings, and one Grid/List switch governs both — Ideas lay out as a grid of cards, or as a list grouped under the video each idea came from.

Each card's checkbox now sits at the head of the meta row rather than among the action buttons, and a checked card shows a blue border. The bulk selection toolbar leads with a clear control and the count of selected moments. Discover's sidebar entry uses a telescope icon, distinct from Search's magnifying glass.

Discover now stays visible when the flags service cannot be reached. A brand's sidebar entry, Discover page, and chapter actions remain available even when the flags dashboard is temporarily unavailable.

The board modal now sizes itself to the moments it holds rather than filling most of the screen. Moments sit directly under the header, and the board's "…" menu shares one row with the moment count, visibility toggle, and Select all.

Compilations

Compilation rows can now be reordered by dragging. Each row has a grip on its left edge; drag it to place the row anywhere in the list in one move. A row with its trim picker open carries the picker along when dragged. The grip is also keyboard-operable: focus it, press Space or Enter to pick it up, move with the arrow keys, and press Space or Enter to drop, or Escape to cancel.

Fixed the transcript in a compilation row's trim picker growing to fill the page. The picker now holds a fixed height: the search field stays at the top, the transcript scrolls beneath it, and Set in, Set out, and Keep this range stay at the bottom.

Fixed pressing a word in a compilation row's trim picker triggering a build. A word press now only moves the playhead and selects the line for Set in and Set out; pressing Enter in the transcript search no longer submits the build form either.

Video Editor

The selection toolbars in the timeline now sit below the track they act on rather than floating above all tracks. The section being edited — its reorder grip, its trim handles, and clips that occupy only the upper half of a track — stays fully reachable while the toolbar is up. The toolbar fades while a trim handle is dragged so the handle's own readout stays clear.

Fixes

Each chapter row in a video's Info tab now shows its timecode, title, and expandable summary — the board action has been removed from this read-only surface.

September 17, 2026

Video Editor

Classic is the new default style for new videos and templates — white outlined captions and a solid white hook text box — replacing Casual. Existing videos and templates are unchanged.

A saved caption or overlay style with no preset recorded now keeps its saved values instead of being replaced with the plain default.

Captions in Hindi, Marathi, Nepali, Sanskrit, Khmer, and Burmese now break where sentences end. Those scripts use no period, so every sentence ran into the next until the word budget or a pause happened to split it; the danda, khan, and pote ma are now treated the way the period is in every other language.

Applying a polish to a video that is showing a different transcript — a translation or a copy — no longer reports success while the editor shows nothing has changed. The polished trims apply, the words from the other transcript stay as they are, and the editor says why.

Clips & Content Library

Videos up to 75 minutes can now be opened in the editor and rendered for publishing. The previous limit was 40 minutes.

A video now takes one processing run and one polish run at a time. Two requests that arrived together used to start two runs and charge the brand twice for the same work; the second request now joins the run already in progress. The reprocess control says when a run is in flight, and the retry button follows it rather than adding another.

The content score no longer shows a "Reach · Watch · Stay" count line above the pillar breakdown. Those category names appeared in the breakdown's own section headers a few pixels below, carrying a different number each time; the score now explains itself through the ring and the pillars alone.

Fixed the content score showing three empty "No top-creator patterns for this category yet" sections on every uploaded video. The explanation is now the pillar graphic on every score. The breakdown's labels have also been tidied: "Hook overlay" is now "Hook text", the caption and cover draw have their own Metadata section, and Engage appears only where it is strong.

The add-label control in the asset details pane now appears on the Info tab as well as the Content tab.

The Clip Ideas page's empty state and footnote no longer say to upload a video between 2 and 20 minutes long. That range is no longer what the page enforces, so both now name the check a creator can act on: a video appears here when no clips have been cut from it yet.

New brands open onto updated sample content. The demo video arrives with four generated clips filed under it, each with its own title, description, content score, and draft posts; the video carries chapters, scene boundaries, cover suggestions, and a content score of its own. Each sample clip shows its portrait thumbnail in the content list.

Cue

Cue can now delete a video's transcripts. Naming a language shows a confirmation card for that transcript; asking without naming one shows a pick-list of every transcript you can delete, marked with the one the video reads today. Cue explains which transcripts must stay — the original, any one a post or reframe reads, and the last one left — and names the transcript it will switch to if you delete the current one.

Cue can now switch which transcript a video reads. Name a language and the video moves to that transcript; ask without naming one and the assistant shows the options and switches to whichever you pick. A language the video has no transcript in is answered with what it does have, alongside an offer to translate.

Cue's import error messages are now specific about what went wrong. A folder or share page on Dropbox, Google Drive, OneDrive, WeTransfer, or Google Photos is recognized and answered with where to get the direct file link instead. A link that is not public, has moved, or points at an unresponsive host each says which of those applies. A Dropbox link to a single file now imports directly.

Calendar

Fixed the Calendar showing posts whose schedule or publish attempt had actually failed. Both the month grid and the list now require the post to be in the published or scheduled phase, not just a date column being set.

Fixed the brand Calendar page going blank for brands with a large post history. Its empty-state check now stops at the first match rather than counting every post the brand has ever had, which timed out on brands with a long history. A background refetch that fails terminally no longer stays silently on stale results, so the "no posts yet" state is never shown for a brand that has posts.

Fixes

Uploading a file that fails now says what went wrong. A rejected file names its type and the formats we accept; the failed-upload list keeps that reason on screen after the toast goes; and a video the backend could not read says so on its row and in the activity list. Videos whose format the browser does not recognize — commonly .mkv and .flv on Windows and Linux — can now be uploaded. A cloud storage import that cannot be started now reports it instead of quietly dropping the row.

Every failed video in the activity list now says why it failed, not just the ones that failed to process. A failed import points at its source, an entry still offering the retry button names Retry, and both now carry the reason under the label rather than the label alone.

Forms no longer go dead after the server turns a submission down. The rejection previously disabled all buttons in the form — including the close button — so a form with nothing to edit had no way forward except to reload the page.

Dates and times that showed "Invalid Date" now render correctly. A timestamp whose UTC offset is written as hours alone — +00 rather than +00:00 — now parses correctly on every surface, and a date or time that cannot be read at all renders blank instead.

September 10, 2026

Clip Ideas

Clip Ideas has been rebuilt around the ideas themselves. The ranked list of videos and the per-video "Mine ideas" button are gone. Two actions mine across the whole library: "Ideas from popular videos" mines the ten most popular videos that have no ideas yet and shows the count and credit cost before you press it, and "Ideas about…" takes a topic and mines the five videos that best match it. A time filter scopes both the mine and the feed to the last 30 days, 90 days, 12 months, or all time.

Asking for a topic now answers twice. The clips you already made about it appear in about a second, while moments nobody has cut yet arrive behind them over the next couple of minutes — so anything already in your library is found alongside what is not. Both answers stand on their own: the library lane says plainly when nothing matches rather than appearing only when the sweep comes back empty.

Each moment card has its own thumbnail of its first frame, set beside the hook and linked to the source video. Pressing the thumbnail opens a preview over the page, where handles nudge either end of the moment by up to ten seconds, and "Make this clip" sends the window as it stands. A window you moved is used exactly when you build the clip; an untouched window still snaps to sentence boundaries.

Pressing a sweep or a topic search now shows a spinner and says what it is doing. Once the run starts, a banner follows it through your videos; moments arrive one at a time and the feed holds a place for the next one so the layout does not jump as each lands. A finished run says what it found, including when it found nothing.

Previewing a moment opens a window over the page instead of unfolding under its card. The feed is two cards wide, so an unfolding preview used to stretch the card beside it into a tall empty box and squeeze the player into half a column.

Cards no longer carry a tier badge. Every video was labelled either a hidden gem or a banger, and the second was really "none of the above" wearing the page's loudest label. The attribution line says what is actually known about a video's performance instead.

Cue's get_clip_ideas now answers "what should I clip next?" with the same idea cards the Clip Ideas page shows — each moment's hook, title, summary, and source video, ordered by how clip-worthy it is.

Clips & Content Library

Content Score is now on for everyone. The Score column and its sort in the Content table, the score ring in the video details pane, the ring on clip-builder cards, and the Analyze tab in the video editor are all live.

Every column header in the Content table — About, Score, Last activity, and Created — is now a button that sorts the list, and the header shows which column is active and in which direction. Sorting by Score ranks a video by its content score, falling back to the clip-evaluation score, with unscored videos at the end in either direction. Clips sit under the video they came from, ranked among themselves, whatever the list is sorted by.

Pressing a chapter band on the clip gallery's source timeline opens Generate Clips with the director's note already scoped to that section — its title and its clock bounds — and the notes section expanded. A video whose analysis found no chapters now shows a "Default section" band spanning the whole runtime, so clip generation is reachable from the gallery for every video.

The clip gallery's per-batch source timelines are replaced by one timeline pinned above the whole gallery. It stays in view through the full scroll, carries a marker for every clip in every batch, and pressing a marker opens that clip's preview wherever the clip lives.

The gallery's "Generate more" button is removed. Generating from the gallery now goes through a chapter band on the source timeline, which scopes the run to that section. The gallery header reports any clip run that is under way — started from Cue, the video's details pane, or the gallery itself.

Hovering a clip pairs it with its place on the source strip across the whole page, not just within its batch.

Clips sharing a content score in the Content table now fall back to the clip-selection score for their order, so the stronger clip leads instead of whichever was made most recently. The clip gallery already ranked them this way, so the two surfaces now agree.

Asking for more clips from a clip now produces clips from the original video it was cut from, not a cut of a cut. Nothing prevented it before: generate_clips, plan_clips, and build_clips each accepted a clip whenever it was at least 60 seconds long, and a shorter clip was refused for being short — which sent the assistant looking for a different video rather than the source.

Fixed the chapter bands on the clip gallery's source timeline being drawn on a narrower axis than the clip markers above them. A band stopped 12 pixels short at each end, so a band covering the whole video read as shorter than the clips did, and every band sat slightly right of the footage it named.

Fixed Combine failing with a validation error when the selection held a clip the gallery no longer shows. It now combines the clips that are visible, and each selected card's number counts through those.

Fixed a clip rejection in the clip gallery that cleared the whole combine selection. Rejecting a clip now leaves every other selected clip checked and numbered as it was.

Fixed the clip gallery and a video's details pane disagreeing about whether a clip run was under way. For as long as the two answers differed, a chapter band could offer a press that Generate Clips then refused, or report a run that Generate Clips did not recognize.

Fixed a clip with unreadable source timing vanishing from the gallery's source timeline without explanation. The timeline now says how many clips it could not place, so the timeline and the clip grid agree on how many a batch has.

Video Editor

A floating toolbar appears over the timeline whenever a section is selected, naming Duplicate, Copy, Paste, and Delete, each with a tooltip for its keyboard shortcut. It is where section-level actions live from now on. The Duplicate button that sat beside Split in the timeline control row is removed, and Cmd/Ctrl+D still works. The toolbar shares the same band as the scene selection bar, so the editor carries no reserved empty strip above the timeline when nothing is selected and the video preview is 48 pixels taller.

Timeline sections are now reachable from the keyboard. Tab moves through them and Enter or Space selects one, which brings the toolbar up; a second press puts the selection away. Each section reads its place in the running order and where it starts in the source video to a screen reader. When one of the editor's floating toolbars closes, focus returns to the surface the selection was made on — the timeline for the section and scene bars, the transcript for the word bar — so Tab continues from where you were working rather than from the top of the page.

Each section's badge now has a tooltip saying what it measures. Resting the pointer on a section in the source view reads "Play order: 2 of 4"; in the edit view, "Source start time: 01:42".

Seven duration and position displays now roll into an hours field past an hour. A search result, a scene tooltip in the timeline, a clip plan in Cue, a clip choice, a cut proposal, a publish cover chip, and an insights duration each showed minutes past 60 on a long video — 62:05 instead of 1:02:05.

Timecodes now read as M:SS everywhere, so a position 22 seconds in reads 0:22 in every place it appears. The clip builder showed a clip's start as 00:22 and a chapter tooltip a few pixels away as 0:22. Cue's frame cards also stop carrying milliseconds they had no use for.

Fixed the editor's Duplicate action duplicating the section playing under the playhead when nothing was selected. A press intended for the browser — Cmd/Ctrl+D to bookmark the page — put a second copy of that section into the edit, and a 13-second video could publish as 27 seconds of the same footage. Duplicate, Copy, Paste, and their keyboard shortcuts now act only on the selected section.

Fixed the video editor's timeline sections climbing over the toolbar that floats above them. An edit cut into many sections could stack them high enough in the z-order that one of them swallowed presses aimed at the toolbar's controls.

Fixed Escape clearing a transcript word selection when the press belonged to something else — a text field being edited, an open dropdown or menu, or a tooltip.

Cue

Cue's transcript now folds a run of tool calls into one collapsed row — "Used a tool" or "Used 5 tools", with "Using…" while calls are still running — instead of stacking one badge per call. Expanding the row lists each tool; one called several times shows as "readResource ×5" and expands to one line per call. Opening or closing a "Thought process" or tool row now animates its height.

Cue can order a brand's video list by upload date, last activity, content score, or title, in either direction, so asking for your best-scoring videos or the ones you have been working on gets the answer without paging through the library. The ordering matches the Content table's, so the two agree on what "recent" and "highest-scoring" mean.

The publish approval card in Cue now works. Its Approve button had been reporting "This card can't publish" on every post while Cue then finished the publish from the chat, because the server that runs a tool call is built fresh for each request and never sees what the client declared when it connected. Cue now restates its capabilities on each call, and the card receives the approval it needs.

Cue now tells a creator what actually went wrong when the AI gateway rejects a turn. A request the provider refused as malformed used to answer "Something went wrong, please try again" — advice that could never succeed since the same request fails the same way every time. It now says the misconfiguration is on our end and that the team has been alerted. A conversation that outgrows the model's context window gets its own answer, telling the creator to start a new thread.

Cue's clip plan review card, video frame cards, and video analysis score cards are now available through any connected MCP client, not just the in-app chat. Video frame cards now show real frames at scene changes rather than frames picked at random, and clicking one opens the video editor at that exact moment.

Searching for documentation from Cue now returns each matching page's title, a highlighted snippet, and a browser link alongside the result — instead of raw titles and malformed links — with an explicit empty result when nothing matches. Searching for images returns each result's filename, pixel dimensions, and the video it belongs to, with paging for large result sets.

Fixed the line breaks in every instruction, description, and prompt Cue reads. Text written across two source lines reached the model with the break still in it, so a sentence could arrive as two fragments on separate lines. The scheduling guidance Cue reads before picking a time zone also arrived as a single run-on line with bullet markers left inline.

Publishing

The post editor's publish-failure banner now shows the reason the publish failed — pulled from the job itself — with a link to the job details, instead of only a generic failure message.

A scheduled post now publishes the video as it was last edited. Editing a video that has a scheduled post re-renders each of those posts individually, so the video that reaches the platform is the one the editor shows, with its own hook text and trim applied.

A post no longer publishes after its creator was told it had not. When the request that starts a render timed out, the publish was reported as failed while the render it had already queued went on to publish the post anyway — and a retry could send it out twice. Both publish surfaces now pre-generate the render job's row ID and claim it on timeout, so a render that was started is visible to the abandon check rather than running silently.

A render whose server died mid-flight now goes back into the render queue for a second attempt instead of failing outright, so one crashed container no longer decides whether a post publishes. A video queued for a publish render no longer gets skipped on every pass through a busy queue.

Fixed several paths where a post was left waiting on video work nothing was doing: approving a publish in Cue, saving an edit on a video with a scheduled post, and approving a publish from the post page each now check that the request preparing the video answered with the job it was asked for, and report the post as unprepared if it did not.

Fixed a failed publish approval overwriting a publish time it did not set. The cleanup now acts only on the publish time the approval itself wrote, and checks whether a publish is already queued before putting a cancelled one back.

Fixed the assistant naming only the scheduling attempt — and not the video — when it abandoned a publish for an already-scheduled post. The message now names the video as the reason nothing was published, and then says what the post's schedule is showing.

Fixes

Documentation search now ranks results by content words. A natural-language query like "how do I schedule a post to YouTube" ranks the relevant guide first. Common function words no longer count as matches, and word stems match across forms, so "scheduling" finds "schedule".

The cookie consent dialog no longer keeps reappearing after you accept it. The consent manager split one decision across two browser storage values, and a half-written pair left it nothing to restore, so it asked again on every page.

A tab that finds a new version of the app now reloads itself at most once per deploy. If the page is still stale after that reload, it shows the "A new version of Radical Tempo is available" toast instead of reloading again.

The Content page now shows one loading state and then the content, instead of cycling through up to five render phases — empty area, skeleton, content, skeleton again, content again — on a visit with no sort in the URL.

Fixed the collapse animation on expandable panels. Their contents now stay visible while the panel folds up, rather than disappearing as soon as it starts to close.

Fixed some animations on the brand creation page.

September 8, 2026

The select-all checkbox on the Content page now reads as checked whenever every visible row is selected, no matter the order they were ticked. A single click then clears the selection.

The bulk actions toolbar on the Content page now appears and counts correctly when clips are selected. Delete, Duplicate, Archive, and Add to apply to every checked row, clips included.

September 2, 2026

Video Editor

The video editor's source view draws the whole video on the timeline. Every frame sits on the timeline strip itself, dimmed and faded wherever the edit drops it, and each trim section is an empty window onto that strip rather than a box carrying its own frames. The playback-order view is unchanged.

A section that covers the same frames as another collapses to a bracket taking half the height of the timeline — the same box with its top or its bottom side dropped — so the frames underneath stay visible and both sections keep their handles. An edit that replays one span in the middle of it reads as a pair of brackets among boxes, rather than as a timeline that changed shape from end to end.

Every section in the source view carries a number. That view draws each section at the frames it was cut from, so the order the sections sit in is not the order they play, and nothing on the timeline said which was which. The numbers read 1 2 3 4 across an edit that plays in the order it was cut and 1 3 2 4 across one that does not.

The source view's layout holds still while an edit is reshaped. The 2 rows a repeated span is split across are worked out once, when the view opens, so adding, deleting or retiming a section leaves every other section on the row it was already on. A trim no longer rearranges the timeline under the pointer.

A section on the edit-view timeline is badged with the time it starts in the source video — 00:00, 00:02, 10:14 — instead of its place in the order the video plays. That view already lays the sections out in playing order, so the old badge only ever counted 1 2 3 from left to right and repeated what the timeline showed. Where each section was cut from is the part that view leaves out.

A section's number stays against the timeline's left edge for as long as any part of that section is in view, instead of sitting at the section's start and scrolling off with it. Reading which section is under the playhead no longer means scrolling back to where that section begins.

The timeline's view toggle sits at the left of the controls row, next to undo and redo, with its Edit option first. It used to sit at the right end, ahead of the seek and frame-step buttons, where a narrow window left it competing with them for space. The row also lays its 3 groups out in columns, so every control stays readable down to the narrowest window the editor draws.

The timeline no longer switches between the source view and the edit view on its own, and it opens in the edit view. Duplicating a span that then plays twice, opening an edit that already replays a span, and carrying a section along the running order with the keyboard all leave the view where you put it.

A cold open's teaser fades to black and to silence over its last quarter second, then cuts to the body at full brightness and full level. The fade covers the captions and overlay text along with the picture, and it shows in the editor preview, in every clip and post preview, and in rendered videos. It ran for a full second before.

A video framed as a facecam over screen content — the screen on top, the webcam below — renders with the height split its framing declares, in the editor preview, in exported videos, and in the timeline thumbnails. Such a layout collapsed to an even 50/50 split before, whatever split it was cut with. Automatic text placement follows the same bands, and the classic split and single-crop framings are unchanged.

The editor's Templates section is a single "Save as template" action for a brand with no templates yet. It used to open the Design tab with an explanatory card that took more room than the Layout and Overlays controls below it, and the explanation now sits in a tooltip on the button.

A timeline thumbnail's decode stops when the editor stops needing it. A thumbnail that scrolls out of view, or that a zoom or a trim drag replaces with one at another size, used to run to the end, decoding every frame it had been given for a thumbnail nothing would draw. A browser lends out only a few hardware video decoders, so each abandoned decode was one the thumbnails you are looking at had to wait behind. Measured on a 79-second video: 6 seconds after a 10-step zoom, the timeline draws all 24 of its thumbnails where it drew 1.

Two timeline sections that draw the same source frames at the same size share one set of decoded frames instead of each decoding its own, so duplicating a section fills the copy in as fast as it can be drawn. Decoded frames also outlive the section that asked for them by 5 minutes, so a section that scrolls out of view and back, or a zoom that is undone, redraws from what is already there.

Every timeline strip reads that same cache, including the strips on a timeline that does not zoom. Such a strip used to hold one entry for its whole row of thumbnails, keyed by the exact fractional size it had measured, so the eighth of a pixel a settling editor moves a row by threw the row away and decoded every frame in it again. Measured on an 8-minute video whose timeline draws 13 thumbnails across 2 sections: that movement used to cost 13 further decodes and now costs none.

Thumbnails appear as they decode rather than a whole row at a time, and a frame that fails to decode leaves its own gap instead of pulling every later thumbnail in the row left into its place.

The editor timeline and the cover picker encode each frame they decode straight to a binary image, instead of going through a base64 string the browser parses back, so both hold fewer copies of every frame and spend less of the main thread. A browser that cannot encode a JPEG returns a PNG, and such a frame is now named and typed after the image the browser actually produced, so a cover set from a video frame in that browser is stored as the image it really is.

Fixed the timeline thumbnails on iOS, where the video they are decoded from was handed to the system fullscreen player instead of being read frame by frame.

Fixed the timeline thumbnails asking the video for the same frame twice in a row, which happens when a section holds fewer frames than its strip has thumbnails or ends past the end of the video. Each such frame is read once, and a strip whose frames the video could not supply in full retries rather than drawing short.

Fixed a single frame that fails to decode blanking every other thumbnail decoded alongside it. A strip keeps the frames that decoded and shows a gap only where one genuinely failed.

Fixed a thumbnail staying blank when its strip was measured mid-transition. A strip under half a pixel tall has no whole pixel to draw on, and it now waits for a size it can draw at.

Fixed the source view's timeline seeking to the wrong place on an edit that replays a span. A click used to move the playhead to the first time the clicked frame plays, whatever it landed on, so the later of two plays could not be split at all. A click now seeks to the occurrence it landed on, and a drag holds the section it started on rather than following the pointer into whichever section it passes over.

Clips & Content Library

The Content table's Created and Latest Update columns are replaced by a single Last activity column, and the list sorts by it. A row says what someone last did and when — "Edited 3h ago", "Published 2d ago", "Uploaded Jul 3" — so the videos being worked on sit at the top instead of wherever their upload date put them. Anything older than a week shows its date, and the tooltip carries the full timestamp, the date a scheduled post goes out, and the date the video was added.

Working on a clip raises its source video too, so a video and everything cut from it surface together.

Rows in the Content table are source videos again, each carrying all of its clips, and a page counts source videos rather than every asset. Filters pick which source videos appear, and a family matches when the source video or any of its clips does — so filtering to Reels still finds the videos whose clips are Reels, and every clip of a matching video stays visible instead of being thinned to the ones that matched. The dimmed "From …" header that stood in for a source video missing from the page goes with it.

The clips under a video in the Content table are ordered by their content score, highest first, so the clip worth reaching for leads the list. A clip the scorer had too little signal to grade sorts last, clips sharing a score keep their previous newest-first order, and the "#N top pick" badge is gone. Each row's score arrives with the row itself, so the score column fills in as the table paints rather than a moment later.

Each card in the clip gallery carries the same content score ring the asset pane shows, in the bottom-left corner of its thumbnail. Hover it for the summary card, click it for the full sub-score breakdown. A clip with no settled score shows no ring, and the gallery reads every clip's score in one request, so a page of badged cards costs the same as an unbadged one.

The clips in a batch are ordered by that score, highest first, with the clip quality score breaking a tie. A combined highlight stays pinned to the top of its batch, and an unscored clip sorts below every scored one. The star and the gold highlight that marked a clip as a top pick are gone from the card and from the preview it opens: they ranked clips by a different measure than the score now shown beside them, so the two disagreed on the same card.

The clip gallery's source timeline and its cards are paired in both directions. Hovering a marker rings the clip's card, the same way hovering a card lights its marker, so a chapter of the source video leads straight to the clips that came out of it. Pressing a marker opens that clip's preview, the same one its card opens, and Enter and Space reach it too.

A batch's title, its preview controls, and its source timeline stay pinned to the top of the gallery while its cards scroll past, so the timeline is still there to read at the bottom of a long batch. Only the batch you are reading stays pinned, and it gives way as the next batch arrives.

A combined clip draws one marker per highlight it was built from, each at that highlight's own place in the video, so a combination of 2 moments 10 minutes apart reads as 2 markers 10 minutes apart. It used to draw a single marker at the very start of the source video. Hovering or focusing one names which part of the clip it is, and gives that part's own start time and duration.

The clips card on a video reports how many clips there are — "8 clips" — instead of splitting them into how many are waiting for review and how many were accepted. Rejected clips are left out of the count, so it matches the gallery the card links to.

Every clip a video produces appears in the library and counts toward that chip. A clip had to be accepted to show up there before, so a clip the pipeline did not pick as a top clip could be generated and never seen. Accepting is gone from the clip gallery, both the check on a card and the Accept button in a preview; rejecting a clip still works from either place and still takes it out of the library.

A clip that opens on its hook says so. A small mark sits after the clip's title, in the clip gallery and in the details pane beside the video list, and pointing at it says what it means and how long the tease runs — "Opens on the hook — a 2.7s tease, then the clip in full". The mark reads the clip as it plays right now, so an edit that takes the tease back out takes the mark with it, and an edit that keeps the opening keeps it.

You can choose which line a cold open teases. "Open on the hook" lists the lines the suggestion considered — the one it picked, plus up to 3 runners-up — and "Use this opening" plays whichever you select. Picking a different line changes only the tease, and the rest of the clip, along with the hook text written for the opening, stays as it was. The suggestion checks only its own pick against the video, so selecting another line swaps that confidence for "Scored from the transcript only", which is what is actually known about it.

The opening suggestions in the Polish tab list every opening you can pick, and only those. An opening this clip cannot play no longer appears at all — it used to sit in the list under an "Unavailable" badge with a dead button.

An opening the suggestion stage ranked last stays in the list and stays pickable, carrying a "Not recommended" chip, its own score, and one sentence saying what the stage held against it — "The tease gives away the payoff", "This clip reads better opening naturally". Picking one was always allowed, and now the button says so.

"Start later" says how far into the clip it starts — "Starts 3.9s into the clip". The number used to be read off the underlying video file, so a 37-second clip could be told its later start was at 21.2s, a moment nowhere in what it plays.

Clip generation uses the hook-unit planner for every brand, which selects each clip's opening and its overlay text together.

Fixed the clip gallery freezing on a video whose recorded duration is far longer than the footage. The source strip's ruler drew one tick for every interval across the whole duration, so a duration in the millions of seconds asked the browser to render millions of ticks. The strip now picks its tick interval from the duration, which also makes a long video's ruler readable: a 3-hour video is marked every 5 minutes and labeled every 30.

Fixed the Content page's detail pane growing so tall for a vertical video that the performance section fell off screen. The video stops at a square, centered with dark space to each side, and anything 1:1 or wider still fills the pane's width. The pane's loading skeleton matches that square and carries a placeholder for the "Added" date, so the pane settles into one height instead of resizing twice while the video loads.

Fixed the status icon appearing on the Source strip in a clip's details pane, where it crowded the marker on a strip too narrow to hold it. The wider strip in the clip gallery still shows a status icon on hover.

Cue

Cue can correct a note it saved about a brand. A correction replaces the note it names instead of leaving 2 saved notes that disagree, and the saved text comes back on the answer, so a client with no way to read a resource can still confirm what a note now says. Removing a note stays something you do in the app, and Cue asked to forget something says so rather than answering as though the note were gone.

The brand-wide post status board answers again. Asking what is scheduled this week, or what drafts are waiting, returned a generic error on every brand large enough to matter: counting one brand's 2,316 posts took 4.5 seconds against an 8-second limit, so the read was cut off and failed with it. The same count now takes 21 milliseconds, and every read of a brand's posts is faster for the same change.

A completed background task says what it produced. Checking a task's status once it finishes returns the result beside the status: the images a cover generation created, the videos a clip run or an import created, the post a publish sent along with its public URL on the platform, or the video a processing or reframe step changed in place. Cue used to finish the wait with a status and nothing else, and then had to guess which of a brand's images or clips were the ones it had asked for.

A task that failed or was cancelled carries no result, so a failed publish reads as a failure with its reason, never as a publish that succeeded and produced nothing.

Asking Cue where to read about something returns a card of matching documentation pages, each with its section trail and the passage that matched, and a click opens the page. Cue could search the documentation before but had no way to hand the pages back as anything you could open. Reading the documentation to answer a question is unchanged, and neither action needs you to be signed in.

Publishing through Cue works the same way in every client. A client that renders no approval card and no confirmation dialog gets the proposal as text along with an approval Cue carries back: you say go ahead in the conversation, and the posts go out without leaving it. A link to each post in Radical Tempo is no longer the only way through for such a client.

That approval is bound to the exact proposal you were shown and to the publish time, and it lasts 30 minutes, so a schedule cannot be answered with a different hour and an approval cannot be reused a day later. Cue picks how to ask from what the proposal came back with rather than from what it believes its client can render, so a publish no longer stalls on an assistant waiting for a decision on a card that was never on screen.

Cue no longer quotes a proposed clip window's timing as the finished clip's. Approving a proposed window unedited is the most common way through the two-phase clip flow, and the build treats such a window as a region to search rather than one to cut, so a window offered as 0:00–0:44 shipped as a 63-second clip. Cue now reads the finished build's per-window outcome before it names a clip's timing, and it describes the longest-clip control as what it is: a target the clip aims near, not a ceiling it is held to.

A YouTube livestream link imports. A youtube.com/live/… URL failed with "Failed to fetch video info" because the import reads a video's ID out of watch, youtu.be, embed and shorts links only, and every accepted YouTube link is now sent in its canonical form. Cue names the link shapes it accepts and says that a stream still in progress has to finish before it can be imported.

Importing a video from a URL checks every redirect the URL takes, not only the address handed in. A URL is refused up front when it points at a loopback, private, or cloud-metadata address, but the request that read its type and size then followed redirects on its own. Each hop is now checked the same way, and a redirect to a refused address, to inline data, or through a chain longer than 5 hops stops the import with a message naming what was blocked.

A URL that accepts the connection and then never answers no longer holds an import open. The check, the redirects it follows, and the image read that follows them share one 30-second limit.

Fixed interactive cards misbehaving inside a chat client's sandbox. Their stylesheet was requested from the wrong place and returned a 404 before loading correctly on a second try, and the cards reported a security warning to the client on every render.

Saving and correcting brand notes works again for all AI models. Adding correction support placed a constraint in the tool's schema at a level some models reject, so those models refused every conversation that included the tool — even when no note was going to be saved or changed.

Publishing

A post no longer publishes an older render when the render it needed could not be started. It published whatever it already had rendered, which could come from an earlier trim, caption set, hook, platform override, or renderer version, so you saw one video in the editor and a different one on the platform. Such a post is now left unpublished with an error and keeps the schedule it had, and publishing it again re-attempts the render, so a post whose render failed once can go out as soon as rendering works again.

A publish no longer fails on a long video because the request that starts its render timed out. That request waits 30 seconds, and the endpoint it calls read the whole composition, the crop and the transcript before answering — a read that grows with the length of the video, so the longer the video, the likelier the publish was abandoned before its render had even been queued. The endpoint now answers as soon as the render is queued and does that reading in the background.

A post is no longer published twice when its render started but the request saying so did not come back in time. A publish checks for a render already running for the post before it decides the render never started, and leaves that render to publish the post when it finishes. When that check cannot be made at all, nothing is published and you are asked to try again in a few minutes. Both the post page and Cue's publish approval do this.

Releasing the cover picker's scrubber on another frame before the first decode finishes cancels the capture you moved past. A browser lends out only a few hardware video decoders, so the slot the abandoned capture was holding goes to the frame you are waiting on.

A frame that never produced an image is no longer offered as a cover. A cancelled capture and a decode that failed both leave a cover with nothing behind it, and such a cover stayed in the picker and in the "Generate" tab's "Start from" strip, where choosing it could only report an error. A cancelled capture is silent, because you caused it by picking another frame, and a decode that genuinely failed still says so.

Fixed a publish failing outright when one of its posts had no readable video, which took down every other post in the same publish.

August 30, 2026

A post no longer publishes its untouched source file when a render was supposed to run first. Any transient failure of the request that starts the render was enough to cause it, the 30-second timeout that request carries included, and the post then went out at the full source duration and aspect ratio, with none of the trims, crop, captions or overlays it was set up with. A post now publishes as it stands only where the render check says so, and one with nothing rendered is left unpublished with an error rather than publishing the wrong video.

Fixed the render check measuring the wrong video for a clip. It judged the clip's full-length source, which can be hours long where the clip is seconds, so a clip could be judged too long to render and then published as its raw source.

Fixed publishing failing on a video built by combining highlights. Publishing reported that the video could not be prepared, and no post of that kind reached the point of being rendered. A crop for a combined video carries position and size but no confidence score or primary-subject marker, and both were treated as required.

August 26, 2026

Cue

Cue can now search a brand's content. It runs the same ranked search the search page runs — spoken words, on-screen text and events, titles and summaries together — and each hit carries the moments the query matched, so Cue can name a timestamp without reading a whole transcript first. Every hit says whether it is an upload or a clip, and a clip names the video it was cut from.

A hit that matched on meaning rather than on exact words now carries a time window to seek to, with the transcript text inside it. A naturally phrased question — "government borrowing costs drive everything else in the economy" — used to rank the right video first and then hand back no moment at all.

Cue reads a video's transcript in pieces, so a long recording no longer exceeds what a single answer can carry. It can also read what a video shows rather than only what it says, and read the chapters that tile a video's runtime, which is the cheapest answer to where in a video something happens.

A transcript now names who is speaking. Each read binds every voice to the person the analysis detected, with that person's name and how confident the match is. A voice nothing accounts for is reported as unmatched rather than filled in with the nearest plausible name.

Cue can take a video file from the machine it is running on. It checks your brand's size, storage, and credit limits before anything starts, sends the file in parts so an interruption costs one part rather than the whole upload, and processing begins the same way it does for a video uploaded through the app.

Cue can answer questions about a brand's whole publishing schedule — what is scheduled this week, what drafts are waiting, what went out last month — instead of reading each video in turn. Each post says whether its platform is actually connected, so Cue does not offer to publish to a platform your brand has never linked.

You can publish and schedule through Cue from a client that cannot show the approval card. Cue picks how to ask from what the client supports: the card where it renders one, a confirmation inside the same exchange where it does not, and otherwise a link to each post in Radical Tempo to approve there. Publishing now always requires an approval Radical Tempo issued, so a client that never showed you one cannot publish on your behalf.

Assistants whose clients cannot browse Radical Tempo's resources can reach a brand's videos, transcripts, posts, memories, collections, and jobs directly. A new brands index lists every brand you can reach and the platforms each one can publish to.

An image layer can be scaled up to 200% of the frame's shortest side. Asking Cue for a badge bigger than the frame's short edge was refused outright, and a layer you had sized up could not be removed and added back — Cue reported its size faithfully and then rejected that same size. The editor's Scale % field accepts the same range, so a size Cue sets is one the editor keeps.

An image layer can be made fully transparent. Asking Cue to fade a watermark out completely was rejected, and a layer already faded to nothing in the editor could not be removed and added back. Cue also knows that fading an image to nothing is not how you take it off a video, because it leaves an invisible layer behind — ask Cue to remove the layer instead.

The editor's offset fields and the rendered video now agree on how far an image layer can sit from the edge it is anchored to. A layer stored beyond the ±50% the fields show opened with a reduced offset in the editor, rendered at its full offset in the video, and had the reduced value written back on the next save, none of it reported.

An image layer no longer disappears when one of its numbers is unusable. A layer whose size, opacity, offset, corner radius, or draw order had been stored as empty was dropped from the editor and the render together, and because the editor then saved without it, the layer was gone for good. Each of those numbers now falls back to its resting value, so the layer keeps its image and its position.

Cue no longer recommends other products when you ask for something Radical Tempo does not do. It states the limit in one sentence, points at the way back in — make the video elsewhere, then import the finished file — and describes what Radical Tempo does with the footage once it lands. Cue will still name a tool you brought up first.

Cue speaks in the terms the product shows you rather than internal data names, and reports a failure by naming the cause and what to do next — "that upload is missing an audio track" instead of a bare "something went wrong". Error codes, stack traces, and internal IDs no longer reach its replies, and it has dropped exclamation marks and emoticons.

Cue no longer regenerates post covers that were fine. When its cover critique could not run, the failure arrived as wording that read like the images being missing, which is the opposite of what had happened. Cue now says its critic is unavailable and shows you the images to choose between.

Fixed five performance tools — corpus percentile, retention moments, drop-point patterns, within-creator comparison, and cohort retention — failing every call with an internal error, so a comparison could not be had at all.

Video Editor

A cold open's teaser now fades to black over its last second, and the body cuts in at full brightness with no fade-in, so the opening reads as a deliberate tease rather than a jump between two pieces of footage. The fade covers the captions and overlay text along with the picture, and appears in both the editor preview and the exported video.

The teaser's audio fades with the picture instead of running at full level underneath a darkening frame. It reaches silence exactly where the picture reaches black, and the next span comes back at full level on the same hard cut.

Dragging a section to a new place in the running order shows the section following the cursor — lifted off the track, shadowed, and slightly transparent — with an empty slot where it came from and a caret marking the boundary a release will land it on. The sections that stay put hold their places for the whole gesture, so the running order no longer shuffles under the caret and the track no longer redraws its thumbnails mid-drag.

Cmd/Ctrl+Shift+Left and Cmd/Ctrl+Shift+Right carry the selected section one place earlier or later in the running order. The section stays selected, so a run of presses walks it the length of the edit. Dragging was the only way to reorder a section before.

A trim handle in the timeline's source view can pass the section beside it, so pulling a section's edge back over its neighbor makes the two share source frames — which is how a span is set to play twice. Dragging up to a neighbor still lands flush by default, and while a handle is held it says what it has reached with a Flush or Overlap badge.

The editor reads a video's chapters from the same place the content details pane and the clip gallery read them, so an edited chapter title shows the same wherever it appears.

The editor requests everything it needs on open in one round trip, rather than in two stages that waited on each other, so it appears sooner.

Accepting a polish or an edit no longer erases a clip's cold open. A proposal now carries back whatever survived of the tease, shortened when a cut trimmed its tail. A proposal that cut through the tease clears it, so a clip that has genuinely lost its cold open reads that way everywhere.

Fixed search result markers landing in the wrong place on a clip that opens with a cold open. Every marker for a moment at or before the teased one collapsed onto the very start of the clip. A moment inside the teased span now scrubs to where it plays in the body, so the conversation around it plays too.

Fixed a drag on the timeline committing its edit to the wrong section. A drag named its section by the place that section held when the press began, so any keyboard gesture that reshaped the running order mid-drag — delete, carry, duplicate, copy, paste, undo, or redo — sent the trim to a different section. A drag now identifies its section by the section itself.

Fixed the drop marker pointing at the wrong gap while a section is dragged to a later place in the running order. It sat one section-length short of the gap it pointed at, so the release put the section somewhere the marker never showed.

Fixed a restore damaging an edit that plays part of the video twice. Restoring a range that covered two cuts folded the repeated sections into one, and restoring a word joined the two places it is heard into a single span — so a video that opened with a teaser and then played the body in full came back playing the body once, with the teaser gone. A restored range is now put back in each place the video plays it.

Fixed cutting a range from the transcript leaving it in place wherever the edit plays that footage a second time.

Fixed cutting or restoring a section re-sorting the whole edit by position in the source video, which turned a video that replays an earlier moment back into an ordinary trim.

Fixed dragging a trim handle saving a different section than the one shown while dragging, on a video whose sections do not play in source order.

Fixed Restore all silence hanging the editor on a video that replays a section.

Fixed dragging a caption word moving it by the wrong distance.

Fixed redo having nothing to put back after an undo, which left the Redo button disabled and its shortcut doing nothing. Undo and redo now walk the same history in both directions, and the redo shortcut does nothing at all once an edit is at its newest state, instead of undoing a further step.

Fixed the timeline's view-mode toggle and the editor both acting on one arrow-key press, which replaced the view you had just chosen. Alt and Cmd/Ctrl with an arrow go back to navigating history and moving the caret, and an arrow key at either end of an edit is passed on rather than swallowed.

Fixed the timeline's shortcut tooltips naming the macOS Command key on every platform. Split, Duplicate, Undo, and Redo show Ctrl and Shift on Windows and Linux, and the Command and Shift glyphs on Apple hardware.

Fixed timeline thumbnails repeating the first frame across the whole strip on a video too short to count a whole frame. The strip is left empty for such a video.

Fixed a chapter whose start or end is not a real position in the video heading the transcript with a timecode reading Infinity:NaN:NaN, or with a negative time. Both are left out, and the chapters around them still show.

Clips & Content Library

Videos in the Content page's details pane display at their native aspect ratio instead of being letterboxed into a fixed 16:9 player. Portrait reels, square, and 4:5 posts are centered on an ambient stage backed by the video's blurred cover image, and wider videos fill the pane's width at their true height.

The clip gallery's source timeline carries a row of labeled bands beneath the clip markers, one per chapter, so a clip's position on the strip reads as the part of the story it came from rather than only as an offset into the runtime. Hovering or focusing a band names the chapter, its position in the video, and the stretch of runtime it covers.

The content table's Latest Update column shows how many posts a video has published — "3 Published" — alongside the publish date.

A video with nothing published and nothing scheduled reads as Drafts in that column, marked with a grey dot, rather than as an em dash. The dash said only that the column had no value, so a library of unpublished work looked like a column that had failed to load.

A Reprocess Asset action in the content details pane's actions menu re-runs processing on a source video. It appears only for a source video that finished importing, and starting a second run while one is still going says so instead of queueing a duplicate.

The content details pane shows a "Reprocessing…" indicator for as long as a re-run is under way, and the menu entry reads "Reprocessing…" and cannot be selected while one is in flight. Both read the run's own job, so they are correct after a reload, in a second tab, and for a run someone else started. A run that has not reported back for a day no longer counts as in progress, so a video whose earlier run was abandoned can be reprocessed again.

The refresh control has been removed from the content details pane. It sat in a full-width row between the tabs and the tab content, where it read as a heading for what was below it, and it was offered on a video whose processing had finished and whose details could no longer change. The pane still follows its video's live updates, and still re-reads every 15 seconds while those updates are unavailable.

Asking Cue for clips now reports its progress on the video. The work splits in two — Cue proposes a plan, then builds the clips once you approve it — and only the all-at-once kind that Create Clips runs was recognized, so the button offered to start a second run on top of the one already going and nothing on the page said anything was happening. The button now says whether clips are being planned or generated and refuses to start another, the run shows as a live status on the video's row and in its details, and it reports progress and failures in the updates panel and the credits breakdown.

Fixed clips built from an approved plan never appearing in the clip gallery. They are grouped under the batch that produced them, showing the format they were built with.

Fixed the Info tab of the content details pane showing stale information. A video selected before its analysis finished kept the summary, format, and category it had when the tab opened — usually "This video has no summary available." — and a person's name assigned by a later pass never appeared at all.

Fixed a change to a video's summary, format, or category looking saved when it had been refused, and fixed a name for a detected person being refused without saying so. The field reverted as though nothing had been typed, and Radical Tempo went on to regenerate the video's metadata from the name that had been refused.

Fixed rows in the content table showing a grey background when the video had clips, which made those rows read as already selected.

Fixed the Created column's dates being a different size from the Latest Update column beside them.

Publishing

The publish entry point in the content details pane sits above the video, so the call to action is the first thing in the pane rather than the last. It reads "Review Posts" while everything is still a draft and "Posts" once a video has scheduled or published posts, matching the heading on the page it opens. The expandable phase breakdown is now a single row, each phase showing a status dot beside its platforms in the same colors the Latest Update column uses.

Fixed an edit or a polish being started on, and applied to, a video that had already been deleted. Every step now refuses, and a refused apply leaves the editor's staged trims alone.

Associating existing content now reports a clear error when one of the selected posts is no longer available, instead of quietly associating fewer posts than were picked, and it is restricted to posts belonging to the brand whose page you started from. A post selected twice, or reachable through two of the brand's platform connections, is no longer reported as unavailable.

Fixed the one-post-per-post-type limit not applying to a post whose platform ID contains a period, which let a second post of the same type be checked and the association then fail on submit.

Insights

The Tempo Insights callout in the post-performance modal has been rebuilt around a takeaway headline and three labeled rows — working, needs work, try next — replacing a flat two-column split where all four strings shared one type size. The takeaway leads, the recommendation sits on a tinted footer row so the actionable line reads first, and the loading skeleton tracks the new row structure so the callout no longer shifts when the insight arrives.

The Content Score breakdown names its packaging-text row after the format the clip was scored as. Title and Hook overlay are measured on opposite formats and share one row, so a short-form clip with no overlay text and no title classification previously read as longform.

The Content Score pill for the weak band is darker, taking its contrast from 2.59:1 to 6.86:1 and putting the weak pill at the same surface lightness as the strong and developing ones.

Docs, Site & Integrations

Shared links to Radical Tempo now produce a proper preview card. The app and the MCP endpoint had no description, image, or social tags at all, so pasting a link to either produced a bare line of text. The share image is about a third of its former size, so previews load instead of being skipped.

Page titles and meta descriptions are shorter, so search results stop truncating them.

Every page declares a canonical address that matches the sitemap, so search engines index the address a visitor is actually served rather than a redirect.

A mistyped or dead link now lands on a styled "page not found" screen in the site's own typography, with a link home, rather than on the bare Next.js 404.

Fixed the Studios page and both legal pages showing the homepage's title and description in their previews, which made them look like duplicates of the home page to search engines. Each now describes itself, and Studios is listed in the sitemap.

Fixes

The Accept invite link in a brand invitation email now lands on the join page. Anyone who already belonged to a brand had no way to accept an invitation: they signed in successfully, landed on their own brand list, and saw no error and no sign of the brand they were invited to. Connecting a platform from a post's editor also returns to the post now, rather than to the brand's platform settings.

Fixed the keyboard focus outline being clipped on the "Filter by" control in the Content page header.

Removed the hover tooltip on the Radical Tempo logo. Screen readers still announce it the same way.

Fixed importing from a link failing outright when the source did not allow the request used to check the file first.

Fixed starting a video upload through an agent failing for every file in a deployed environment. A video whose upload could not be prepared no longer appears in the brand's library as importing forever.

August 19, 2026

Cue

You can now add a logo, watermark, or badge to a video or brand template through Cue. Describe where the image should go — "add this to the top right of my clip" — and Cue places it as a new layer on top of what was already there, visible for the whole video; on a template it reaches every video using that template.

After adding an image layer, you can ask Cue to move, resize, change the opacity of, or rename any layer you already have — only what you asked about changes. You can also ask Cue to remove a layer: it drops from the video or template and leaves every other layer alone, and the image stays in your library. If you name a layer that doesn't exist, Cue comes back with what's there; a layer from a brand's default template is explained as such and offered to be edited on the template, where the change reaches every video using it.

Cue's multiple-choice questions now show visual options. When all choices are images, they appear as a thumbnail grid you tap to pick — single- or multi-select. When an image choice carries a description, it shows as a thumbnail beside its label. When all choices are short clips, they appear side by side as posters you can play before picking — only one plays at a time, and picking is a separate step so comparing never commits by accident.

Long Cue replies — ones that fan out across many parallel actions — no longer stop midway with an error. Cue groups updates while a reply streams in, so a long multi-step turn stays readable and responsive the whole way through.

The Cue conversation list is faster: previews for all conversations now arrive in a single request instead of one per conversation. Each conversation also now shows a preview of its opening message — previously, untitled conversations showed only "(Untitled)" and titled ones had no preview line at all.

The floating Cue button now sits on a white disc with a hairline border and shadow, so it stays legible wherever it's parked — over cards, tinted backgrounds, and video frames. It also moved closer to the screen edge, lining up with the sidebar's icons on brand pages.

Cue no longer refuses to reframe a video on the grounds that it exceeds a platform duration limit when the video is within that limit. Length can arrive as advice alongside the result, but it no longer substitutes for the requested action.

When combining a highlight reel and two clips cover the same footage, Cue now picks the higher-quality one and names the choice, rather than asking you to decide between clips you have no basis to compare.

Several Cue tools — image generation, post cover, title and description, and performance reports — were presenting parameters without their full contents, so Cue had to guess at them or leave them unset. Those parameters now reach Cue correctly. Image generation requests that previously failed because the target argument was treated as optional now succeed, and a wrongly shaped argument shows the exact format to send.

Video Editor

Videos with a chapter breakdown now show chapter headings in the transcript panel — each titled, numbered, and time-stamped — and clicking a heading seeks the video to where that chapter begins.

The scene track in the timeline switches granularity as you zoom. Zoomed in, it shows individual shots; zoomed out past the point where shots become too small to click, it shows chapter bands labeled with their titles. Selecting a chapter band selects every shot inside it, so a whole chapter can be cut in one action; shift- or ⌘-clicking extends the selection across chapters the same way it does across shots.

Fixed the timeline selecting neighboring chapters when you pick one. Shots that span a chapter boundary now belong to whichever chapter holds more of them, so selections no longer bleed into adjacent chapters. The track also keeps showing chapter bands until shots have enough room to be usable click targets, so wide and fullscreen windows no longer switch to shot granularity prematurely.

The Polish tab now appears only on clips. Asking Cue to polish an original video explains why it can't run and offers to generate clips first or trim the video directly.

Clips & Content Library

The "Filter by" entry point on the Content page now appears as an icon-and-text control without an outline or hover fill, matching the size and hover style of the storage callout beside it. The applied-filter chip row sits tighter under the header, and the active-filter count badge is centered in a fixed circle.

The Latest Update column in the assets table now shows a labeled status — Published or Scheduled — with a small status dot, instead of a green checkmark that competed visually with the content score sitting next to it. Videos with nothing published yet show a dash.

The bulk-actions toolbar's action buttons — Delete, Duplicate, Archive, and Add to — now match the weight of the "Selected" count label, with icons scaled to sit beside the text and each icon leading its label.

The loading skeleton for an asset with no collections now shows the "Add to" chip on the title row, matching where it appears in the loaded state. Previously the skeleton placed a separate row below the title, causing a layout jump when the content arrived.

The asset actions menu is now organized into labeled General, Download, and Manage sections. Icons use a lighter weight with a subtle fill, Delete is marked in red, and the submenu chevrons are smaller so they recede.

Publishing

A video's publish-type eligibility is now judged on the output duration of the edit, not the source footage. A recording trimmed down to shortform length is immediately offered YouTube Shorts, Instagram, TikTok, and Facebook Reels without waiting for a render; a video trimmed below a minimum is offered shorter formats instead. Speed changes count too.

Downloading a reframed variant now applies the variant's crop, aspect ratio, and fill settings, matching what publishing it produces. Previously downloads rendered at the base video's framing, so they were not a reliable preview of the published output.

Opening the post editor for a video whose upload or import never completed no longer shows a "page not found" error — it returns to the video list with that video selected, the same way a video still processing does.

Insights

The content score breakdown has been reworked. The overall score now appears in the modal heading, each pillar's description moves into a tooltip on its heading so the three pillar scores read as a set, and repeated "Edit video" / "Edit post" links collapse to a single "Fix" per row.

A shared grid across all sub-score rows fixed three alignment defects: a sub-score of 0 no longer shows a zero-width bar indistinguishable from an unscored row, a row with no fix link no longer stretches its bar past its neighbors, and the "Early signal" tag no longer forces its label to wrap.

The post-performance modal now holds a stable height when you switch between comparative and cumulative views or change the platform filter.

The metrics table in the post-performance modal fills the available width evenly across platform columns, each metric row shows its icon, long values such as watch time stay on one line with the unit rendered smaller and muted, and the cumulative totals sit in the first column aligned with the comparative view.

Docs, Site & Integrations

The "Built for" section on the marketing site has been redesigned from a card grid to a tabbed layout. The four audiences sit in a list on the left and the selected one expands into a panel on the right with its description and highlights, each audience's accent color running through its label, icon, and the panel's checkmarks.

On mobile, the demo section's landscape recording and vertical clip now stay in sync throughout. Mobile now uses the same clip timeline as desktop — tapping a clip on the timeline picks the pair and keeps it loaded for as long as you watch.

Fixes

An app crash that showed a blank "Application error" screen — triggered by changing a filter or page while another was still loading, or by a platform connection delivering data before the browser finished navigating — has been fixed. Recovery previously required a page reload.

Importing from Google Drive no longer fails with an OAuth scope error for accounts that also have YouTube connected. If the Drive picker fails to open, it now shows an error instead of leaving you on a blank Google page.

August 12, 2026

Cue

Cue conversations can now be archived instead of deleted. The floating panel's action becomes Archive — one click, no confirmation, since it is reversible — and the panel lists only the conversations you keep. The full Cue page's rail gains an Active / Archived filter: active conversations archive in one click, and archived ones can be unarchived or permanently deleted after a confirmation.

Cue can now cancel background tasks it queued — clip generation, imports, edits, and publishes — as long as the work has not yet started. Asking it to "cancel that" or "just stop" cancels the task; once work is under way, Cue says so plainly rather than claiming it stopped something it did not. Cancelling a publish attempt leaves the post scheduled; Cue says so rather than implying the post was unscheduled.

Publishing a video to several platforms at once now shows a single card — a per-platform toggle list, a shared timing control, and one confirmation. Approving publishes each selected platform and reports the outcome per platform, including partial failures. The card keeps its per-platform status across a reload, and the retired summary lists each platform's real outcome — Published, Scheduled, or Publishing… — rather than "Already handled."

The publish card's timing is now an interactive control: you can switch between publishing now and scheduling, and pick a wall-clock time shown in your time zone. Approving forwards the chosen mode and time; the default tracks what Cue proposed.

A post that is already scheduled can now be published immediately or rescheduled to a new time from the Cue publish card, without first cancelling the existing schedule manually. If publishing fails after the old schedule is cancelled, the original schedule is restored; the card says if that restoration also fails.

When you ask Cue to preview the clip plan before it renders — or give specific instructions about what moments to feature — Cue now runs a free planning pass first, shows the candidate windows in a review card, and builds only the clips you keep. Unchecking a window removes it from the batch; declining the set cancels the request.

The brand and video shown on the current page are treated as defaults rather than the subject of every request. Naming a different video — "add captions to my Q3 launch video" while looking at something else — makes Cue find the one you named instead of acting on what is on screen.

Cue now reports when it cannot read a video, post, or saved note, instead of silently working from what it already had.

Proposal cards — publish, polish, and edit approvals — now appear as the last item in Cue's turn, after any introductory text.

A failed Cue turn now shows an error inline in the conversation, in the same place the response would have appeared. Reopening a thread where a turn failed shows the same error where it happened, rather than the conversation simply stopping.

Answering one card in a Cue turn no longer disables the other cards in that turn. An answer sent while Cue's response is still streaming is held and delivered once the turn settles, rather than being dropped.

Applying an edit or polish proposal now refuses if the video's trims have changed since the proposal was made, rather than silently overwriting the intervening manual edits. The card asks you to request a fresh proposal.

A stalled edit analysis no longer makes a video permanently uneditable — Cue can propose a new edit while an earlier analysis is in progress. Cue also now looks up an edit's actual status before describing it, so it can no longer report an edit as still running when it has already finished.

Cue now sees each video's duration and clip eligibility before deciding which to use. When the video you most likely meant cannot be clipped — for example, a sample clip under the one-minute minimum — Cue names it and its length, identifies the nearest eligible video, and asks before switching.

A background task's completion update now arrives in the open Cue conversation even if another tab started announcing it and then closed mid-way. The hold is re-checked when it lapses, so the update is delivered rather than waiting until the next page load.

Clips & Content Library

The Content page's header controls have been redesigned. The Active / Archived visibility toggle is now a dropdown beside the "Content" title, and the five standalone filter dropdowns — Type, Status, Source, Collections, and Created Date — are consolidated into a single "Filter by" button that opens a step-by-step filter builder. Applied filters appear as a count badge on the button and a row of removable chips visible only while at least one filter is active. Filter state stays in the URL, so shareable links and pagination are unchanged.

The clip builder's timeline now spans the full source video duration, so each bar shows where the clip sits relative to the whole video instead of always stretching to fill the available width.

Captions

Captions now break at sentence ends rather than running the start of the next sentence onto the same line. Periods inside words or URLs do not count as sentence ends, and a closing quotation mark does not hide the period behind it. Quotation marks are now read according to the transcript's language — languages that close a quotation with an opening-style mark, as German does, now break correctly at sentence ends — and English captions no longer break after abbreviations such as Mr. or p.m.

Insights

The content score breakdown now reflects the 0.5 engine: Watch, Stay, and Engage pillars replace the previous set, Reach potential appears as a separate early-signal chip outside the score ring, and a pillar with too little data shows "not enough signal yet" rather than zero. Scores from the previous engine continue to render with their original pillars.

The content score breakdown now includes a Title ingredient. A score that contains ingredients the app does not yet recognize from a newer engine is now displayed rather than hidden — the unrecognized ingredients are simply omitted.

The Tempo Insights callout in the post performance modal now holds its shape while the summary loads, so the modal no longer jumps in height when the insight arrives.

Fixes

Downloading a video while a render is running no longer reports a failure if the render outlasts the endpoint's own time limit. The browser continues waiting and finishes the download when the render completes, or surfaces a real error if it fails.

Downloads now stop when you navigate away from the page. A render that fails to start shows the reason the render service gave rather than a generic message, and a repeated polling failure shows a readable message instead of a raw database error.

The video editor's blocking overlay no longer counts time the render spent queued behind other renders against the render's own budget, so it no longer clears before a queued render has actually started.

August 7, 2026

Cue

Brand notes saved in Cue now arrive in every conversation automatically, so a preference stated weeks ago — how titles should sound, who a video is for — shapes the answer whether or not you mention it. A brand with more notes than fit is shown how many there are and where to find the rest.

The floating Cue panel opens with tappable starter prompts instead of a static greeting. The suggestions are chosen from where you are in the app — cover, copy, and scheduling on a post; edits and reframes in the video editor; performance and clip ideas on a video page — and tapping one sends it as the conversation's first message. Each prompt carries an icon in a tinted tile that matches the color used for that journey on the full Cue page.

Cue now understands short phrases that point at what's on screen. Typing "translate this" while a video is open is no longer treated as off-topic — Cue knows when you have a video in view, so a request whose meaning depends on what's in front of you is understood as being about it.

When a brand has used all its Cue credits for the week, an in-chat banner explains that Cue is paused and shows when credits will automatically renew.

The Cue tab in the studio sidebar shows the full-color Cue mark when it's active, and reverts to monochrome when another tab is open.

The floating Cue panel now stays open through clicks elsewhere on the page, focus moving to another part of the app, and Escape. The close button in its header is the only thing that closes it.

Approving a publish proposal to one platform no longer locks the cards for the other platforms in the same turn. Each card tracks its own approval independently, so publishing to several platforms at once works in any order.

The edit card now distinguishes between a run that found nothing to trim and one where the editor found cuts but declined to apply them. The two outcomes are described separately, and neither mislabels the editor's reasoning as what was requested.

Cue's video progress strip now clears in a tab that has been left open past when the underlying work could still be running. A chat open for hours no longer keeps a stale stage label and ticking clock on screen, or keeps polling behind them.

When a model provider is temporarily unavailable, Cue no longer shows a raw error — a clear message appears instead.

Video Editor

The brand logo overlay on new videos is off by default. The layer still appears in the editor's Overlays list with your brand's logo already attached — one click enables it — but the logo no longer appears in videos unless you turn it on.

Clips & Content Library

Combining clips where one covers footage another already contains is now refused with an explanation. A highlight reel built from such a selection would repeat the same footage; clips that merely share a beat at their edges still combine normally.

Browsing an S3 or GCS bucket in the file-import flow now requires you to be signed in.

Insights

YouTube audience-retention curves now start at their real opening value instead of dipping from zero. The combined multi-platform view normalizes each curve to start at 100% so YouTube and TikTok share a common baseline, and the YouTube-only view scales its axis to surface rewatch peaks above 100% with a dashed baseline.

The post-performance metrics table opens in Comparative view by default.

Fixes

The unsaved-changes warning now fires when the browser navigates away from the app through a full-page load, not just an in-app navigation. Edits in the post, template, and video editors could previously be lost without a prompt.

The platform dropdown in the "How your post is performing" header stays on one line even when the post filename is long. The subtitle wraps to the next line instead of squeezing the dropdown.

The Cue chat entry in the brand quota usage panel now shows the correct color.

August 6, 2026

Duplicating a clip from the Content page now works for clips created before the pipeline started recording their source video — previously the action silently did nothing, with no copy and no error. The duplicate appears immediately as a sibling alongside the original, under the same source video, and keeps the original's metadata.

Row checkboxes on the Content page no longer uncheck themselves immediately after a duplicate succeeds.

August 4, 2026

Cue

We've added Cue, a context-aware chat assistant that can automatically perform various tasks, like looking up analytics, scheduling posts, or suggesting ideas for post thumbnails or descriptions based on our knowledge about each platform. We'll be continuing to improve and update Cue over time.

Publishing

A "Use this cover for all platforms" toggle in the post editor's cover widget copies the current post's cover onto every other cover-supporting post of the same video, mirroring the existing "Use this hook for all platforms" switch. Turning the toggle off restores each post's previous cover. Snapchat and X are skipped since they don't use covers; TikTok is included only when the cover is a frame of the video.

Clips & Content Library

Clip picks in the content list are now ranked by their content score, with the clip-evaluation score as a tiebreak, so the "#N top pick" badge and the score shown in the row always move together.

Fixes

The audience-retention tile in the post-performance modal no longer claims a platform "doesn't report second-by-second retention" when the data simply hasn't synced yet. YouTube and TikTok now show "Not available yet" in that case, while platforms that genuinely don't expose retention keep their existing copy. A failed query now shows a distinct error state instead of being mistaken for a sync delay.

The publish placeholder in the asset details loading skeleton now shares a card with the video placeholder, matching the layout of the loaded state.

July 30, 2026

Long videos that previously failed mid-render with a timeout error now complete successfully.

July 29, 2026

Dropdown menus and popovers no longer select an item immediately on open or pick up stray clicks while closing — the animation previously slid the content across its full height, sweeping items under the still-pressed pointer, and now uses a subtle nudge. Side panels use the same nudge in place of their previous full off-screen slide.

July 28, 2026

Video Editor

Videos up to 40 minutes long can now be edited and rendered, doubling the previous 20-minute limit. Longer videos still show the duration limit message.

The quick-edit color swatches in the template and video editors each edit only their own color. Editing one swatch no longer changes another that happened to share the same color, and two matching colors show as separate swatches instead of collapsing into one.

The applied layout template card no longer lights up across its full width on hover, so it no longer reads like the layout template selector above it.

Publishing

YouTube and TikTok posts can save their visibility settings as a brand default. A "Save as brand default" button in the post editor's Settings section stores the current visibility — and, for TikTok, the comment, duet, and stitch permissions — as the brand's default for new posts. The saved defaults can be reviewed and cleared per platform under a new Post Defaults section on the brand settings page.

Saving a post after choosing a large custom cover image works reliably. Custom covers now upload the moment you set them, so the image itself no longer travels through the save request.

Clips & Content Library

The content list remembers your "assets per page" choice. Picking a page size saves it to your account, so returning later — from the sidebar, another brand view, or another device — keeps your last selection instead of resetting to 20.

The pagination footer below the content list is shorter and lighter: the rows-per-page control is a plain "20 per page" dropdown, and the page-number field is a compact square that grows to fit longer page numbers.

Archived videos appear in search results with an "Archived" badge, and clicking one opens the archived view with the video selected — instead of the regular content list, which hides archived videos and made the result look broken.

Studio Navigation

The brand switcher menu has been reworked. The list is headed "Your brands", and the brand you're currently in is marked with a green check instead of linking to the page you're already on. A pinned footer holds a single "Add brand" link — or, when you're at your brand limit, a legible "Upgrade to add more" row with a usage count — and only the brand list scrolls once it grows tall.

The left navigation rail shows the active page with a filled icon, a bold label, and a subtle highlight, and the Content, Analyze, and Connect icons read more clearly.

Docs, Site & Integrations

This What's New page groups each release under topic headings, with every change as its own paragraph, so individual updates are easier to scan.

The ghost clips below the shorts row on the landing page reveal on scroll again instead of rendering static.

Fixes

A newly deployed update no longer interrupts your work. Open tabs show a refresh prompt instead of reloading mid-task, and reload automatically only when it's safe — right after a navigation or while the tab is hidden, and never while there are unsaved changes in the video editor, post composer, or template editor.

The sign-in screen no longer flashes an empty card while loading — the intro animation waits for the sign-in options to render before it opens.

Studio pages no longer intermittently fail with a server error right after a deploy.

July 21, 2026

Video Editor

The Hook Text section has one field per platform, each with its own AI-suggestion carousel. Clicking a platform name previews that hook on screen, and the post page opens each platform directly on its own hook without requiring a "Try suggestion" click first.

A "Use this hook for all platforms" toggle applies one hook everywhere and can be turned off to restore each platform's individual copy. Enabling it from a post propagates to every platform in one save.

End text automatically positions itself to avoid covering faces on screen, matching hook text's existing behavior. It picks a position that balances staying near its configured spot against how much of its on-screen time would overlap a subject's face. An "Avoid covering faces" checkbox on the End Text panel turns this off for a single element.

Auto placement for hook and end text weights the opening moments of a clip more heavily when choosing a position, so a face that appears only later in a shot no longer pushes hook text away from the best early spot.

Manually adjusting the offset for hook or end text turns off auto placement for that element, so the control matches what's on screen. Re-checking the box re-enables it.

Block and Outline are independent toggles for hook and end text, so both can be on at the same time. A background block appears only when Block is on, and it hugs the text instead of spanning the full overlay width.

Long director's notes in the Clip Builder highlights panel clamp to one line with a "Show more / Show less" toggle, keeping the aspect-ratio and speed controls in their pinned position.

The Templates section shows a "Save as template" prompt with a clear explanation when the brand has no saved templates.

Opening a video that has no transcript no longer crashes the editor — it opens normally with a message in the transcript panel.

Opening an asset whose video failed validation shows a placeholder instead of blanking the panel.

Captions

Japanese and Chinese captions apply a character budget — about 14 characters per line at the default setting — instead of a word count, render without inter-word gaps, and fall back to Noto Sans JP or Noto Sans SC glyphs.

Right-to-left captions in Arabic, Hebrew, Persian, and other languages display words in the correct reading order, in both the editor preview and published renders.

Extended-Latin, Vietnamese, Cyrillic, and Greek characters render in their correct glyphs across all built-in caption and overlay fonts.

In multi-speaker videos, secondary speakers are distinguished by hue rotation — a blue preset becomes coral for the second voice — instead of a lightness shift that made colors dull.

The secondary-speaker tint is no longer applied to the primary speaker in single-speaker videos.

The Romantic preset no longer uses a font that was never loaded.

The Ocean, Cartoon, Pop, and Friendly presets no longer show the wrong casing label in the style picker.

Clips & Content Library

The content library groups clips under their source video in every view, including filtered ones. When only some of a video's clips match the active filters, they appear indented under a dimmed "From …" header, and the clips chip reads "3 of 14 clips" to show how many are visible against the family total.

The expand/collapse chevron no longer appears when there are no clips in the current view to expand.

A live "Generating clips…" badge on source video rows shows when generation is in flight, so starting it from a filtered view no longer looks like nothing happened.

Accepted clips can be rejected from the clip card or preview modal, as long as the clip hasn't been saved to a post, scheduled, or published.

The lower-left status widget shows only active file uploads, since processing and import progress already appear in the activity bell.

The activity panel — the bell icon in the studio header — loads much faster for large, mature brands. It fetches only the recent activity it displays rather than every video and post in the brand, and completed jobs older than 14 days no longer appear. In-progress activity always does.

The Generate Clips button degrades gracefully when the asset is temporarily inaccessible, instead of surfacing a raw error.

The setup checklist on the content page hides quietly when it can't load, rather than failing the whole page.

The content page recovers automatically to page one when the URL contains an out-of-range page number.

The uploads list on the content page stays within the viewport and scrolls on short screens.

Asset thumbnails whose source image was removed show the unavailable placeholder instead of an error.

Brands & Connections

The brand connect page highlights platform connections that are missing permissions. A warning color on the info button makes it immediately visible which cards need attention — for example, a TikTok connection where publishing is authorized but analytics requires a separate authorization that hasn't been granted.

The brand Updates popover no longer fails to load for brands with large recently-imported libraries.

Docs, Site & Integrations

Shared links to the Radical Tempo landing page show a branded preview image instead of a fallback icon.

The free-to-use announcement banner no longer breaks its last word onto its own line on narrow mobile screens.

A missing space in the landing page showcase heading is restored.

The documentation pages, docs search, and llms.txt endpoint no longer fail in production.

Fixes

Opening a post for a video that is still processing sends you back with an explanation instead of a "Not found" page.

Previewing a search result at a matched moment no longer crashes on clips whose timestamp had extra fractional digits.

July 16, 2026

The video editor and template editor have an "Avoid covering faces" option in the hook text controls. It is on by default and repositions hook text away from faces; unchecking it pins the text at your configured offset instead. Templates carry the setting forward to their videos, and individual videos can still override it after decoupling the element.

Fixed publishing or scheduling an X post from its post editor, which was failing with a validation error. The issue affected only posts opened directly in the editor.

The crop overlay in the video editor stays in sync with the video at all playback speeds. Playing at non-1× speeds such as 1.25× previously caused the crop mask to appear one frame ahead of the frame actually shown.

The single-frame step buttons (← and →) move exactly one frame at a time and no longer occasionally double-step or stall.

July 14, 2026

Search

Content search is on for all brands. The results page is a two-pane layout — a scannable list of videos and their clips on the left, and a detail pane on the right that previews the selected result without leaving the page.

Each result shows why it matched: an "Exact" or "Meaning" label, the section of the video that matched (title, summary, or transcript), and the matched passage itself.

Clicking a result cues the preview player to the moment the search term appears, so you can watch from that exact point rather than scrubbing to find it.

Search has moved to the top of the sidebar.

The empty search screen shows clearer guidance and example queries, grouped by what you can look for: what's spoken, what's on screen, or what a video is about.

Search results are more accurate for non-English videos.

Video Editor

Hook text automatically positions itself to avoid faces. When a clip has detected face zones, the hook text shifts to a clear area of the frame so it never overlaps a subject while they are on screen — both in the editor preview and in the final rendered video.

AI hook text suggestions are quality-ranked, and the top suggestion appears everywhere by default: in the editor, the post editor, and the rendered output.

In both the video editor and the post editor, once a suggestion is applied the sparkle button becomes previous/next chevrons with a counter, so a suggestion you passed can be reached directly without cycling through the whole list again.

The timeline starts fully zoomed out, so the complete video scope is visible on open.

Clicking outside a selected clip section reliably deselects it.

Cutting or restoring a section no longer moves the playhead to a different word.

Reframe changes appear immediately on the first apply, without requiring a second run.

Long videos over 30 minutes show readable timeline tickmarks in the asset panel and clip builder.

The template editor opens with the first element already selected.

Captions

Multi-line captions with a background box hug each line instead of spanning the full caption width.

Block-style captions no longer paint a thin empty box during silent sections.

Publishing

A Regenerate button in the post editor header refreshes the AI-suggested title, description, and tags for a video, with an optional note to steer the results — for example, naming a guest. Fields you have already edited are never overwritten.

Renaming a detected person in the video details or editor automatically triggers a metadata refresh, so suggestions pick up the corrected name.

Accented and non-Latin text in post titles, descriptions, and tags is Unicode-normalized, so characters aren't over-counted against platform limits and Instagram hashtags with non-ASCII characters highlight in full.

Insights

The per-platform KPI strip includes a "your best video" column showing that video's numbers alongside the top- and bottom-quartile medians.

The "How your videos open" section shows the actual opening frames of each cited example, labeled with their timestamps.

Funnel stages have been renamed from "the click" and "the watch" to reach → the watch → the stay, so the language reads correctly for feed formats where nothing is clicked.

Clips & Content Library

A Retry button appears on failed render entries in the activity panel. Clicking it re-queues the render with the same settings; after three failures within 24 hours the button is replaced by the existing Review link.

Speed options in the Generate Clips modal, batch control, and clip preview modal are consistent at 0.75× – 2×.

The "Generate Clips" quick action is hidden for videos that aren't eligible for clip generation.

Brands & Connections

Connected LinkedIn pages display their real follower count. The count is fetched from LinkedIn when a page is first connected and reflected in the post destination list.

Brand logos in brand settings and the switcher crop to fill their tile instead of leaving white space.

Account & Onboarding

The sign-in page has been redesigned: options appear on a frosted card over the branded background, with phone and email one-time-code forms on a single screen behind a tab toggle instead of separate pages.

An animated intro plays while the sign-in page loads — the Tempo logo appears with rotating captions, then the card expands to reveal the form.

The same animation plays on the /studio entry screen while deciding which brand to open, replacing the skeleton screen.

The account usage popup no longer shows negative credit counts when a brand is over its weekly quota.

Fixes

Returning to a stale tab after waking a laptop no longer shows errors — the session refreshes and the query retries automatically.

Toast notifications have a new dark compact style with faster animations.

July 7, 2026

Video Editor

The zoomable timeline is on by default for everyone. Zoom controls let you expand and compress the timeline to work at any level of detail, and a toggle to the left of them switches back to the legacy timeline. Your preference is remembered.

The timeline is fully virtualized for long videos: thumbnail frames and caption words render only for the visible window, and the transcript panel stays responsive through long recordings.

The timeline ruler steps up to legible tick marks at every zoom level — every 10, 15, 20, 30, or 60 seconds.

The editor preview plays back more smoothly on long videos. Caption lookup and overlay text measurement no longer re-run on every playback frame; they update only when the underlying content actually changes.

Dragging a trim handle shows a live preview of the frame at the cut point you are dragging to. The playhead stays put while you drag and snaps back to its original position when you release.

The Save Updates button shows a persistent "Last saved" time. It reads "Last saved: never" before your first save, remains visible while you have unsaved edits, and survives a page reload — the time only ever reflects a real manual save, not background system writes.

Fixed the "No Crop" reframe option letterboxing horizontal source videos into the clip's existing vertical output frame. Selecting "No Crop" or "Original" now outputs the video at its true source aspect ratio.

The editor preview updates immediately after any reframe change, without a page reload.

Clips & Content Library

Assets can be archived from the asset details menu or in bulk from the selection toolbar. Archived assets disappear from the main library but remain fully intact — posts, analytics, collections, and direct links are unaffected. The Content page has Active and Archived tabs instead of the former "All" view.

A "Source: Uploaded / Imported" filter separates bulk-imported back-catalog content from your own uploads, making it easy to archive an entire import in a few clicks.

The Archived view groups clips under their parent assets, and an archived clip whose parent is still active surfaces as its own top-level row so it stays reachable.

Multiple assets can be duplicated at once from the bulk selection toolbar. Each duplicate is fully reprocessed from the original video file — transcripts, crops, and thumbnails are regenerated rather than copied.

The clip preview modal has a playback speed control. The same picker used in the clip builder, including the recommended-speed label, is available inline, and the preview immediately plays at the selected speed.

News and politics videos now show correctly under News & Politics rather than the wrong category.

The Content page list no longer scrolls back to the top when you select a checkbox or open an asset's details panel.

The Content page shows the "No content matches the current filters" empty state instead of upload widgets when the Source filter returns no results.

The "Generate Clips" quick action is hidden in Source-filtered and Archived views.

The "Add to Collection" bulk action is labeled "Collections" so it fits on a single line.

The "Selected N" label in the bulk toolbar no longer sits slightly higher than the action buttons.

Publishing

The post editor updates in real time. Fields you haven't changed yourself update live as they change — including edits made by the Cue AI agent while the post is open — without a page reload. Your own unsaved edits are preserved, and these background updates never enable the Save button or trigger the unsaved-changes prompt.

Insights

Brand insight reports have a dedicated investor-grade print layout: an editorial cover, an executive summary with the top moves and best-vs-worst growth gap, a contents page, and a method page showing the reach → click → watch funnel. Each platform gets a chapter with top- and bottom-quartile cards, pillar diagnosis, per-second retention curves, and cover comparisons, closing with a methodology page.

Templates

The brand template editor no longer shows hook text as disabled on templates saved before hook text became a template setting, and no longer flips the toggle off on open.

Docs, Site & Integrations

The API reference is part of the documentation site at /docs, sharing its navigation, search, and styling instead of living at a separate /api-docs page.

The Radical Tempo marketing landing page has been redesigned with distinct layouts for the core and studios variants, a new feedback loop animation, and a "Built for" section.

Fixed the mobile video carousel freezing on smaller screens.

July 1, 2026

Fixed TikTok interaction permissions (Allow Comment, Allow Duet, Allow Stitch) being saved inverted. Toggling a permission on and saving now keeps it on the next time you open the post, and vice versa.

Fixed just-published YouTube and Facebook video posts still appearing as eligible to publish in the publish menu. They now correctly show as already published until you make new edits.

Fixed the "Save Progress" button and unsaved-changes warning sometimes activating immediately on page load before any edits. System-driven updates no longer mark the form as having unsaved changes — only real user edits enable saving.

Fixed saving a post failing with "Failed to save changes to publication." when no destination account was selected. Clearing a connected account from a post now saves correctly.

June 30, 2026

Insights

Brand insight reports have been significantly redesigned. Each report opens with a cross-platform band: a headline, a per-platform upside card showing how much each platform's best videos outperform its worst, a "Start here" digest featuring the single highest-impact move and a ranked list of the rest, and an orientation panel explaining the analysis. Below the band, each platform appears in its own card under a colored header, split into separate labeled sections when a platform spans multiple formats.

Each platform card leads with an outcome summary and its Content and Packaging pillars, with insights and moves color-coded by funnel stage — "the click", "the watch", or "reach" — and annotated with where to act. Lever pillars are organized as Style, Content, and Packaging.

The method band includes a funnel diagram showing how reach converts to a click and then to a watch, with the lever for each conversion labeled on the arrow and both downstream signals feeding back to earn more reach.

A "What good looks like" section at the top of each platform card places top-quartile and bottom-quartile profile cards side by side, each showing median performance across up to six key metrics, so the gap between your best and typical videos is immediately visible.

The retention, opens, and packaging comparisons are two independent ranked lists rather than forced-paired rows. Each card is self-describing and sizes to its content.

Reports can be downloaded as a clean, paginated PDF — the app chrome is hidden, cards don't split across page breaks, and each platform section starts on a fresh page.

Clips & Content Library

Clips generated from combined highlights show a "Combined" badge on their clip card, with the merge icon and the number of source clips.

The Generate Clips speed picker always includes the recommended playback speed, even when the suggestion falls between the standard options such as 1.1× or 1.3×. The recommended speed is inserted in order, labeled, and pre-selected, and the same speeds are available in the clip builder's batch speed control.

The clip builder's framing dropdown changes only the crop within the 9:16 output. Picking "Short (3:4)" or "Square (1:1)" letterboxes that crop inside the 9:16 frame rather than changing the clip's output shape, so the gallery, preview modal, and editor all agree on a clip's shape.

The preview modal sizes its canvas to the clip's actual output shape, so any mismatch is visible without opening the editor.

The top navigation progress bar no longer flashes during clip generation. Background refreshes in the clip gallery — including real-time updates as new clips arrive and the refresh after queuing another batch — no longer trigger the bar.

Video Editor

Fixed logos and image overlays drifting downward when a clip is reframed to a shorter aspect ratio. Overlay positions are now measured as a fraction of the frame height and width, so they stay in the same relative position across all output aspect ratios.

The playhead stays on the same source frame when you finish dragging a trim handle, instead of jumping to a different one.

The "Save Updates" button is disabled after saving and only re-enables once you make a new edit.

Brand logos no longer reappear after being removed from a brand template.

Image overlays whose source image was deleted no longer appear in the editor preview, matching the rendered output.

Publishing

Fixed publish links in the asset details panel sometimes pointing to an incompatible platform format — for example, linking a short clip to a Facebook video instead of a Facebook reel, which caused an "Incompatible" error. Publish links now always target a format the video qualifies for.

Fixed the post page briefly showing the video's first frame instead of its cover image before the video is played.

Fixed published videos occasionally rendering without asset-level crops when the post had a trim-only edit.

Fixed per-post composition settings such as overlays and captions being silently dropped from the rendered output.

June 25, 2026

Fixed an inconsistency in the publishing popover where the same post could flip between publishable and "Missing details" depending on which other post was open alongside it. All posts are now evaluated with the same logic — a post with a resolvable destination and all required fields is consistently shown as eligible and can be published without saving it first. TikTok posts still require the checkboxes be filled out before they can be published.

June 23, 2026

Publishing

The publishing popover has been redesigned. Ready-to-publish posts appear at the top of a single sorted list; platforms that still need details or a connected account appear below, dimmed, with a status label and a hover shortcut to connect them. Already-published and already-scheduled posts move to footer rows.

Scheduling is a two-step flow with an explicit confirmation step before the date is applied, and a backdrop dismisses the popover on click.

Clips & Content Library

Clip generation surfaces Radical Tempo's recommended framing and speed settings in the Generate Clips modal, labeled "Recommended" so the suggested option is clear at a glance.

Clip speed is a selectable setting when generating clips.

The entire Performance section in the asset details panel is clickable, making it easier to open the performance details view.

Video Editor

Caption speaker colors tint every word from a secondary speaker, not just the word currently being spoken, so you can tell who's talking throughout the entire caption. The most-active secondary speaker receives the strongest contrast tint, making the two dominant voices the easiest to distinguish.

The brand's default template shows a "Default" badge in the template selector, consistent with the badge already shown in brand settings.

Docs, Site & Integrations

Links to the Trust Center are available on the sign-in page and in the main site footer.

Fixes

The style preview palette for hook text and end text no longer shows a focus-color swatch — that color applies only to captions.

Fixed the cookie consent overlay getting stuck in an unresolved state.

June 19, 2026

Publishing

Posts no longer require a separate approval step before publishing. A post is publishable as soon as it has a connected destination and the platform's required fields. The "Approved" and "For Review" states have been removed.

New videos come with a post pre-filled with AI-suggested title, description, and tags, ready to review and publish with no manual setup.

The post editor shows a "Last saved…" time next to the Save button. Publishing counts as a save.

The "Edit Cover" frame picker previews frames in the post's actual cover crop — portrait-cropped and subject-tracked — so scrubbing shows exactly what each frame would look like as a published cover.

Covers on posts without a custom reframe use asset-level subject-tracking, matching how the video player already works.

The video preview in the post editor is consistently sized across all platforms.

Fixed the "Last saved" indicator in the post editor briefly showing a future time, such as "Last saved in 4 minutes", immediately after saving.

Templates

Brand templates can be duplicated or deleted. A star in the editor marks the brand's default template, and each template card in brand settings shows Edit, Duplicate, and Delete on hover.

The first template you create for a brand automatically becomes its default, and deleting a template leaves existing videos untouched.

The brand's default template previews automatically in the video editor and shared video player for uncommitted videos, always showing the latest version of the template. Saving, downloading, or publishing locks in the template at that moment.

Applying a brand template carries over its end text content to the video.

Hook text stays per-video and can be edited while the element is synced to a template — styling, timing, and layout remain template-controlled, but the copy is yours.

The template editor's color pickers show your brand's color swatches for one-click color application, matching the video editor.

Brand template previews show a generic "Hook Text / Example" placeholder instead of the current video's hook text, and the template editor no longer includes a hook text input.

Template preview cards no longer show placeholder hook text, caption, or end text pills for templates that don't define those elements — a logo-only template previews just its logo.

Opening a brand template and immediately navigating away no longer triggers a "leave without saving?" prompt on an untouched template.

Custom image overlay timings saved in a template are preserved correctly when the template is reopened.

Logo images appear only during their configured timing window in the template preview.

Template thumbnails in brand settings render each template's configured font.

Video Editor

The Reframe controls have moved from their own tab to a collapsible Layout section in the Design tab, with the current aspect ratio shown in the collapsed header.

Turning templates on in the video editor immediately applies a template rather than revealing a grayed-out selector.

End Text defaults to the vertical center of the video, so it no longer overlaps captions when both are enabled.

Dragging trim handles no longer freezes the editor or blacks out the preview.

Insights

Insight reports include a "What retains audience" section listing the moments that keep viewers watching, alongside the existing retention cliff analysis.

Brand insights shows a separate report tile for each platform and format, so YouTube Long-form and YouTube Shorts surface as distinct entries.

Entity cards show accurate screen and talk time. People identified primarily by appearance rather than voice are counted and ranked by screen time, so the main on-camera subject is no longer sorted below minor talkers.

The People section is visible for assets where names were inferred with high confidence.

The performance widget in the asset panel shows numbers consistent with the Performance pages.

Brands & Connections

Brands with a completed channel audit have a Playbook view, reachable from brand settings, showing exactly what Tempo learned and what it will apply to generated content. Entries are grouped by surface — metadata, hook text, covers, clip duration, and theme — with a confidence rating and reasoning for each, and YouTube Long-form and YouTube Shorts show separate, sometimes opposite, recommendations.

Style Guides renders your brand's shared channel voice, do's and don'ts, and brand terms alongside per-platform measured habits such as title length, hashtag count, and emoji rate.

Brand colors can be set and edited with a visual picker — a saturation/brightness area, hue slider, and hex input — so you no longer need to know a color's exact hex value to add or change it.

Clips & Content Library

The Reels filter in the asset library categorizes clips by true orientation and length: vertical clips under three minutes, including 3:4 portrait, go to Reels, and everything else goes to Long Form.

Each batch in Clip Builder has a format dropdown in its title row to choose the output aspect ratio, updating the crop preview across every clip in the batch at once.

Importing files from cloud storage no longer briefly shows a duplicate asset in the activity widget during the handoff between import and processing.

Fixes

Fixed a crash when navigating to a brand after it was deleted or your access was removed.

June 11, 2026

Video Editor

Hook text, end text, and captions each have their own independent style preset. Choosing or customizing a style for one element no longer changes the others, and recoloring one element's palette stays isolated to that element.

The video editor pre-fills the default image overlay with your brand's logo, so you no longer have to add it manually each time.

The reframe preview accurately shows how clips will be cropped, matching the final rendered output.

The style-preset selector renders its preview pill in the font, color, and outline of the active style — whether a saved preset or a custom style — so the pill matches what you see in the video.

Templates

Brand templates no longer apply automatically when you open a video. Videos now open with no template selected unless you choose one.

The video editor shows when the template applied to a video is out of sync with its source. An update notice lets you pull in the latest changes; videos already up to date show a "Synced with template" confirmation.

Brand settings lists video templates as thumbnail preview cards, each rendering the template's own layout, with an Edit action on hover.

New brand templates are automatically named Template 1, Template 2, and so on, starting from the highest existing number.

Caption word-count settings defined in a template carry over to any video the template is applied to.

Insights

Brand insight reports render as a structured in-app dashboard — with KPI summaries, gap comparisons, retention charts, lever cards, and more — when full audit data is available. Reports without structured data continue to display as before.

Brand insight reports include a "Download PDF" button that produces a clean print view of just the report content.

The brand insights hub shows per-platform reports alongside the unified view.

Docs, Site & Integrations

A link to the site status page is available in the user menu and support menu. Docs and status page links also appear in the footer of the Radical Tempo marketing site.

The MCP server provides an interactive editor nav card — showing a video's thumbnail and label — that links directly into the video editor, for AI tools and integrations that support MCP widgets.

Fixes

Fixed a blank loading state in the details panel on the Content page.

May 28, 2026

A documentation site with search and structured guides is available at /docs.

An llms.txt endpoint is available for AI agents.

---

New entries are added before each weekly release.