Changelog

Two caption blocks could briefly overlap: the line you were reading lingered for a moment after the next one had already appeared, so for a fraction of a second both were on screen at once. In continuous speech this was happening about a tenth of the time.

The old line now clears exactly as the new one arrives. Where the speaker pauses it still lingers a moment, as it always did, so the last word does not blink out.

A video was being called mixed content whenever it was cut between cameras often — which is exactly what a well-shot interview or podcast does. Every automatically detected video in our library came out as mixed or action; not one was recognised as speech, including interviews that talk from start to finish.

How busy the cutting is no longer affects that decision at all; only how much the picture itself moves. An interview cut between three cameras is now treated as speech, so its clips are the length speech deserves, are chosen by what is said rather than by what moves, and keep the camera on faces.

Picking a type yourself still wins, as before. If the video looks very different from the type picked — Sports on something that talks throughout — the project notes now say so instead of leaving it unexplained.

Finishing a sentence used to be skipped entirely for anything filed as sports, on the reasoning that those clips should end on the edit instead. That worked for a match, but an interview filed as sports lost the protection with it — and those clips stopped mid-sentence.

Every clip now runs to the end of what is being said when the sentence finishes nearby. When it does not — a match where nobody is talking for minutes — the clip still ends on the edit, exactly as before.

Captions that arrive without punctuation are repaired whichever way the video was added, instead of only on the slower of the two import paths.

Clips taken from videos with automatic captions used to end in the middle of a thought — the speaker was still talking when the clip stopped. Those captions arrive without any punctuation, so nothing in them said where one sentence ended and the next began, and the clip had to stop somewhere.

Sentence boundaries are now worked out for such videos before clips are chosen, so a clip runs to the end of what is being said. In a full-length talk show we measured, all five clips finished their sentence instead of none, and each one only needed between a fifth of a second and two and a half seconds more to get there.

Videos whose transcript already has punctuation are unaffected and cost nothing extra.

Split screen used to pick whoever had the biggest face on camera, which in a fight or a match meant a row of spectators filling the panels while the people who mattered shrank. Panels are now built from people who actually speak in the clip, so a crowd shot stays a crowd shot instead of becoming four portraits of strangers.

Clips where nobody on screen is speaking no longer split at all — they keep the full shot, which is what they were worth in the first place.

Three-way splits now ask for a sharp enough source. On smaller videos the clip drops back to a two-way split rather than magnifying each panel until faces go soft.

Sports clips now follow faces in close-up shots instead of always showing the full frame.

Wide play still uses the full frame, so nothing gets cut off while the action is spread out. The choice is made shot by shot, so a celebration or a reaction fills the screen and the play around it does not.

When three people talk in the same shot, the clip can show all three side by side instead of picking two of them. Shots where only two people speak still use the two-way split, so faces stay as large as before.

Who gets a panel is decided by who actually speaks, not by who happens to be on camera. A person sitting silently in the background no longer takes up a panel and shrinks everyone else.

The Activity video type has been removed. We measured it against the same video with and without it and could not tell its effect apart from the normal run-to-run variation, so it was not earning its place on the screen. Auto, Podcast and Sports remain, and projects created with Activity keep working exactly as before.

A project detected as a mix of speech and action now shows "Mixed" instead of naming a type you can no longer pick.

Automatic type detection no longer mistakes a multi-camera conversation for an action video. Camera switches alone used to count as fast editing, so podcasts were treated as mixed content; they are now handled as speech, with clips that follow the conversation.

The first video type is now called Auto, and every project processed with Auto shows which type was detected, for example "Auto → Podcast".

A new "What to look for" field on the create screen lets you steer clip selection in your own words, such as product demos, contrarian takes or every Q&A answer.

Podcast, Activity and Sports each give the AI a short brief about the kind of moments to favor, so the type you choose shapes the selection, not only the framing.

Clips now start at the beginning of a sentence and end at its end. When a clip would have stopped mid-sentence, it runs on until the sentence finishes, a few seconds past your length limit if needed.

Fetching the part of a video a clip needs no longer floods the activity feed. Retries happen quietly, and you see a single line only when the fetch really failed.

Viral scores reach the top of the scale more often for strong videos. Weak sources still get an honest lower score.

Two new video types when you create a project. Podcast shows two people who share the frame side by side and pulls wide when the speaker steps out of frame. Activity is for vlogs and hands-on footage, weighing what is said and what happens on screen together.

The theme control in the header is now a single button instead of three. It shows the theme you are in, and one press switches to the other one. If you want the theme to follow your device, that choice is still in Settings.

The header buttons are in a new order: theme and language first, then your plan and Open Studio, with your account button at the end.

Your account button now sits at the end of the header row, after the language and theme buttons.

The language menu now shows proper flag images. On Windows the flags were showing up as two-letter boxes, because Windows has no flag characters built in — they are pictures now, so they look the same on every device.

The FAQ answer about Starter and Pro was still describing the old Pro limit and did not mention 4K. It now matches what the plans actually give you.

Pro now renders up to 4K. Your clips keep the quality of the video you started from, so a 4K source comes out in 4K instead of being capped at 1080p.

The Pro plan now includes 100 source videos per month.

The language button in the top bar shows just the flag, without the country code next to it.

Projects are kept for 30 days after you create them, then deleted automatically along with their clips.

You get a warning on the project for the last three days, counting down to the day it goes, so you have time to download anything you want to keep.

The whole interface is now available in Arabic, including right-to-left layout, picked from the language switcher in the top bar.

The whole interface is now available in Thai, picked from the language switcher in the top bar.

The whole interface is now available in Japanese, picked from the language switcher in the top bar.

The whole interface is now available in Simplified Chinese, picked from the language switcher in the top bar.

The whole interface is now available in Korean, picked from the language switcher in the top bar.

The whole interface is now available in Russian, picked from the language switcher in the top bar.

Deleting your account used to ask you to type the English word DELETE, even on non-English pages. It now asks for the equivalent word in your language instead.

The whole interface -- landing page, panel, settings, everything -- is now available in five more languages, picked from the switcher in the top bar.

If your browser is set to one of these languages, Joynubi now opens in it automatically.

The language control in Settings is gone -- the switcher in the top bar is now the only place to change it.

Languages we don't support yet are easier to spot in that switcher: a clear "Coming soon" label next to a full-color flag, instead of the whole row looking faded.

The short caption example on clip cards ("This is the best part") no longer gets translated with the rest of the interface -- it's sample content, not interface copy.

The language switcher now sits in the panel's top bar and the home page header, next to the theme switch, not just in Settings.

It also previews the other languages we're planning to support, marked "Coming soon" until they're ready.

The interface can now be switched to Turkish from Settings → Appearance, next to the theme switch. English stays the default.

Privacy Policy and Terms of Service stay in English for now on the Turkish site, pending a final review of the legal text.

Light, dark and system theme can now be switched from every screen, not just the home page and Settings.

The Server cleanup card now has a "Run server cleanup now" button; the server picks the request up within a minute and the card shows the result.

Server clean-up now keeps only the most recent 8 GB of Docker build cache instead of everything from the last week, so it actually frees space after a busy deploy day.

Russian was removed from the content languages along with Arabic; the caption typefaces cover Latin-script languages only.

Three new caption styles, Steel, Candy and Circuit, replace Classic, Box and Minimal. Clips that used a removed style switch to the closest remaining one the next time they are rendered.

Captions now sit slightly lower in the frame, still above the area platforms cover with their own controls.

Russian captions are drawn with a typeface that includes Cyrillic, so every style renders correctly for Russian content. Arabic was removed from the content languages.

The Team page shows each member's last activity and whether Patreon is linked; the manual add-account form is gone because sign-in is Patreon only.

Six new caption styles: Neon, Coral, Slab, Paper, Arcade and Soft. Each uses a different typeface, colour palette, highlight and letter case, so the twelve styles are easy to tell apart.

Paper is the first light style: dark text on a white bar with a violet highlight.

Style preview cards now show the bar colour of boxed styles and follow your content language when changing letter case (Turkish dotted and dotless i).

All new typefaces were checked to include every letter of the twenty supported content languages.

Content language now offers Czech, Danish, Dutch, Finnish, Hungarian, Indonesian, Italian, Malay, Norwegian, Polish, Portuguese, Romanian and Swedish, alongside the existing seven.

Only languages whose letters every caption style can draw were added, so subtitles render in the style you picked rather than a fallback font.

Creating a project now saves its settings (aspect ratio, layout, clip length and count, caption style, language, video type, hook text, clip mode) as your defaults for the next one. Projects you already made are not changed.

The new-project screen now starts from your saved defaults for video type, mirror, upscaling, layout and hook text; before, those always started from the factory values.

Settings gained Layout, Hook text and clip-mode defaults.

The clip player remembers your volume and mute across clips and page reloads.

The Maintenance page now checks first and lists what can be removed, with the space it would free; nothing is deleted until you confirm.

Download folders left behind by an interrupted import are now cleaned up too.

A weekly server-side clean-up reports Docker build cache and old benchmark files on the same page; it only measures until it is switched on.

The weekly clean-up is guaranteed to be scheduled exactly once.

Manual clips now detect the camera cuts inside the chosen range, so framing and split screen are decided per shot instead of once for the whole clip.

A clip that stayed in split screen or the whole-frame view from start to end could fail to render. It renders now.

The AI Clipping / Manual clip switch is now a highlighted toggle, so the active mode is obvious.

Caption styles are shown as preview cards in the new-project screen, the same cards you see in the clip editor.

The separate "Don't clip" list is gone; in Manual clip the range you enter is exactly what gets cut.

The new-project screen now has two tabs. "AI Clipping" picks the strongest moments as before. "Manual clip" takes a start and an end time and turns exactly that part into one clip, with captions, framing and your look applied. No length limit; quick buttons set a 3 minute Shorts or 5 minute Reels range.

In Manual clip you can list parts inside the range to skip; the clip jumps over them.

Processing timeframe and Don't clip moved from the AI screen into the Manual clip tab.

The home page footer now shows a Changelog link with the running version next to it.

Re-rendering a clip (for example after changing the caption style) keeps split screen and the whole-frame view. Before, only the very first render could produce them.

The sidebar now shows a Changelog link with the running version underneath it.

The text at the top of a clip now shows for the first three seconds and then disappears, instead of staying on screen for the whole clip.

When a shot is a wide or general view (nobody close to the camera, or no face at all), the clip now shows the whole picture in the middle with a blurred top and bottom, instead of a tight crop of an empty corner.

The "Follow the subject" option is hidden for now. It was not actually tracking anything yet and quietly behaved like automatic mode. It will return once subject tracking is measured and switched on.

Split screen now also works when the source and the clip are both landscape: the two people are shown side by side.

Switching an already rendered clip to Split screen now re-scans the frame for both people. Before, the clip could come back as a single crop with a note saying no two people were found.

New "Split screen" layout: when two people share the frame, the clip shows both, one above the other (side by side in landscape). Pick it per clip or as the project default.

The new-project wizard now offers all four ratios, including 4:5, and all four layouts.

Layout names in the clip editor are clearer: Automatic, Centred, Fit whole frame, Split screen, Follow the face, Follow the subject.

Split screen panels are now shaped correctly for square, 4:5 and landscape outputs.

If split screen is not possible for a clip, the clip's note says why instead of silently falling back.

A person could sit half outside the frame for the first second after a cut. Cuts are now placed on the exact frame, so the camera re-centers immediately.

Switching between the split view, the full picture and the close-up is now a short cross-fade instead of a hard cut.

When the person in focus changes within a shot, the camera glides over instead of jumping.

Clips that switch between people many times could fail to render when the split or full-frame view kicked in. They render again.

Each new shot now re-centers on the speaker's face, so people no longer sit at the edge of the frame with an ear cut off.

Switching between people at a cut is now a short, smooth move instead of an instant jump.

When two people share the frame, the clip shows both, one above the other. When the speaker leaves the frame, the whole picture is shown until they are back.

A stray detection can no longer steal the frame from a clearly visible face.

Captions now show fewer words at a time, so each line is easier to read on a phone. Measured on production clips, roughly half as much text is on screen at once.

A short title now stays at the top of the clip the whole way through, instead of only during the opening seconds.

Loudness is evened out, so quiet passages no longer drop away on phone speakers.

Clips found from replay data now get a written title and hook instead of a raw line lifted from the transcript. On recent production clips, five in six were missing that copy.

The hook no longer repeats the title. It used to be filled in from the title whenever nothing else had written one, so the same sentence appeared twice.

The most-replayed moment still keeps its place in the results exactly as before. Only the words describing it changed.

Vertical clips now zoom in further on the speaker. Measured across production clips, the face takes up about a third more of the frame than before, and the zoom now kicks in on every clip instead of most of them.

On a few clips with a small or distant subject, the camera follows a little more actively as a result. Everything else is framed exactly as it was.

Vertical clips are framed on the speaker again instead of being fitted whole into a blurred background. Measured across twelve clips, the face takes up about 29% more of the frame, and on the clips that suffered most it more than doubled.

The framing now follows whoever is speaking, and it sees smaller and farther faces than before.

Clips also finish several times faster, because the framing no longer runs twice.

On longer clips the vertical framing could take more than five minutes, and the result was discarded even though it had finished. Those clips now keep the framing that was produced for them.

The same fault kept the machine busy on work nobody received, so queues move faster now as well.

When the framing engine cannot improve a clip, the clip now falls back to the standard vertical crop right away instead of waiting minutes for a result that gets discarded. Busy queues clear noticeably faster.

The project activity log now says when clips were chosen from replay, motion and audio signals instead of by reading the transcript, so a run that had nothing to read is no longer silent about it.

When the person speaking changes, the vertical frame now moves with them instead of staying on whoever it started on.

Moments where two people are both talking can be shown one above the other, so neither of them is cut out of the frame.

Emphasis in the audio is matched with a small, slow zoom rather than a fixed crop for the whole clip.

For a short window this afternoon, new videos could not be processed: transcription and face tracking were unavailable, so no new clips were produced. The cause has been fixed and processing is running normally.

A clip rendered earlier today took several minutes longer than normal to finish. The experimental step responsible has been switched off, and render times are back to where they were.

Until now, a camera path was worked out once and kept forever, so improvements to framing only showed up in brand-new clips. Re-render an older clip and its camera path is now rebuilt with the current version, which means yesterday's fix for the wandering crop applies to it too.

Nothing is recomputed in the background: the work happens only when you re-render a clip, and it adds a few seconds to that render.

When someone turns away or the shot changes, the crop used to drift towards nothing for a second or two — you would get the top of a head, half a face, or a pair of shoes. It now stays with whoever is actually on screen.

Across a set of real clips, the share of frames with nobody properly in view dropped from about one in twelve to about one in a hundred.

Rendering a clip is about a quarter quicker, so the whole batch reaches you sooner.

Finished clips are roughly a fifth smaller, which means less waiting when you download them and less to upload wherever you post them.

The person in a clip is now framed towards the middle of the picture instead of drifting to the edge. Faces that used to sit half cut off against the side of a vertical clip are kept inside a safe centre band.

Vertical clips now look for a face in far more videos. Until now only footage recognised as a plain talking head was followed by face; anything busier followed movement instead, which often left the camera on a microphone or a hand while the speaker sat outside the frame.

Frames where nobody is visible at all are down by about two thirds.

The Team page opens again. It was failing to load every time it was opened.

You can now set a processing timeframe on a new project: give a start and an end, and only those minutes are downloaded and processed. It sits right in the main settings, and it uses fewer credits than processing a whole video.

Video type, colour look, mirror and upscaling are now remembered. Set them once in Settings and every new project starts with your choice — you can still change any of them for a single project without changing what you saved.

Hook text is no longer a permanent text box. It shows what you have written, with an Edit button when you want to change it.

Suggesting hook text with AI now fills the field in for you instead of leaving you to copy the suggestion across. The other suggestions are still there to pick from.

Maintenance is now a single Clean up button that removes the temporary files left behind while videos are imported and clips are rendered. It also runs on its own once a week.

Scheduled cleanup, older project media and the disk usage reading have been removed from Maintenance.

The server load chart is now labelled with percentages, and the current CPU and memory figures are shown beside it.

The two CPU readings on that card are now named separately, because they measure different things: one is how busy the processor is right now, the other is how long the queue of waiting work is.

You can now mark parts of a video as off-limits before it is processed. Clips are never taken from those minutes — not even partly — and on YouTube links those sections are not downloaded at all, so processing starts sooner.

The video type now explains itself. Normal picks moments from what is said and keeps the camera on faces; Sports picks moments from motion and scene cuts and follows the action instead.

Maintenance now charts CPU and memory over the last hour instead of showing a single reading, and it updates while you watch.

Maintenance can prepare a nightly report of old project media: how many projects and how much space could be freed. Nothing is ever deleted on its own — the report waits for you, and one button frees the space.

The team page now shows, for each member, when they joined, when they last signed in, how many projects and clips they have, and whether they have signed in at all yet.

Home is now the first item in the sidebar instead of sitting at the bottom.

The privacy, terms and changelog pages now carry the Joynubi wordmark, which takes you back to the homepage.

Ssemble has been added to the comparison on the homepage.

Clip length in settings now uses the same three ranges as the new-project form, so the setting and the form finally agree.

Caption examples in settings are now drawn with the real caption styling — the same preview you see on a clip — instead of small colour swatches.

The watermark position is now a corner picker you can see, and the colour look has its own card.

The suggestion language is now a single switch: on, titles and descriptions are written in the language spoken in the video; off, they are always English.

When several videos are being processed at once, paid plans now get a larger share of the machine. Before, every job in flight competed equally, so a free-plan video could slow a paid one down for as long as it ran. Each plan still gets a real share — nothing is starved — and when only one video is being processed it still gets the whole machine.

The pipeline now watches how busy the machine is and stops taking on new work while it is saturated. Work already running is never interrupted. This keeps the site responsive while videos are being processed.

Mirror and sharpening are now on paid plans. On a free account both controls are visible but locked, with a link to the plans — they will not silently do nothing. Your setting is kept either way, so if you upgrade later, the choice you made is still there.

Project cards no longer show the small timeline strip under the clip count.

The new-project form is wider, with more room for each setting.

The viral score now runs on a wider scale. Before, almost every project topped out in the mid-eighties no matter how strong the material was, and the best clip in a weak project looked much the same as the best clip in a strong one. Strong projects now reach the top of the scale and weak ones stay honestly low, so the number tells you something.

The score no longer appears while clips are still being chosen. It used to show up on the first clip and then change on its own a few minutes later — the same clip, a different number — because the score is worked out by comparing a clip against the rest of its project, and the rest did not exist yet. It now waits until every clip is picked, and once you see it, it does not move.

Clips arrive one at a time instead of all at once. Previously the first clip appeared, then the remaining four flipped to "Queued" together and landed as a batch. They are still made in parallel and the total wait is unchanged; you just see them delivered in order.

Hook text is written for every clip. It was only being produced on some videos, so on most projects the setting was on, the plan allowed it, and nothing was drawn — because there was no text to draw.

The new-project screen asks four things: clip length, how many clips, caption style and aspect ratio. Everything else — video type, layout, video language, colour look, sharpening, mirror and hook text — moved into an Advanced section that starts closed. Colour look, sharpening and mirror could not be set at all on this screen before; now they can, and they carry through to every clip.

The new-project form is narrower and tighter, and the platform icons under each aspect ratio are large enough to tell apart.

In the clip editor, "Why this clip" is separated from the settings above it and sits at the bottom. Opening a setting no longer closes it, so you can read why a clip was picked while you change it.

Videos that mix fast editing with a lot of talking used to produce clips titled from measurements alone, like "Hard impact · 00:35". Those clips now read the transcript, so titles, hooks and scores describe what is actually said.

The score on each clip is now shown on the scale you see everywhere else in this space. The number is larger; the order is not. Which clip we think is strongest, and which clips clear the quality bar, are unchanged.

Scoring understands English phrasing. Questions, contrasts and direct address were only recognised in Turkish before, so English videos scored flat regardless of how they were written.

Every clip now shows four bars — Hook, Flow, Value and Trend — with a sentence naming the two strongest. A bar we could not measure shows a dash instead of a number: videos without replay data have no Trend to report, and inventing one would be a guess.

Caption templates no longer share a single typeface. There are three now, and each template keeps its own letter size so the text stays the same height on screen. Two templates also change letter case: one uses sentence case, one lowercase.

The caption template cards show the real typeface, so what you pick is what gets drawn into the video.

Owners have a new Maintenance screen: disk in use, live memory and CPU, leftover files from abandoned uploads, and old project media. Nothing is ever deleted without showing you the list first, and the original video and its audio are always kept.

The hook text panel showed internal placeholder labels instead of real ones. It now reads properly again.

Speaker-change marks from YouTube subtitles (the ">>" arrows) no longer appear in the transcript or get burned into your clips.

Captions sit slightly lower and are a little larger, so they read more easily on a phone.

Picking an AI hook suggestion now sets the clip title to match, instead of leaving you to type it twice.

Colour look and sharpening are visible directly under Advanced rather than hidden behind a section. "Upscaling" is now called "Sharpening", which is what it actually does — it crisps edges, it does not enlarge anything.

Your default colour look can now be chosen in Settings. The value already existed but there was nowhere to see or change it.

Typing a time into the trim fields no longer moves the handle on every keystroke; the value is applied when you finish.

Every panel screen now has a link back to joynubi.com, and the changelog page has a way back into the studio. Previously there was no route home from Settings, Account or Team.

The Pieces section was removed from the clip editor. Clips that were already split still render exactly as before.

On a phone, the buttons above a clip ran past the edge of the card — Regenerate and Edit drifted off the right of the screen and were partly unreachable. They now wrap onto a second line and stay inside the card.

The text drawn over the first seconds of a clip is called the same thing everywhere: hook text. It used to appear under four different names depending on which screen you were on.

Starting a clip stops whichever clip was already playing. Before, the first one kept playing out of sight and you heard both at once.

Clip length is now three choices: under 30 seconds, 30 to 60, or 60 to 90. You used to set the shortest and longest separately, which made it possible to pick a range that could never work.

Hook text can be switched off for a whole project. With it off, nothing is drawn over your clips.

The Advanced options section is gone from the create screen. What you see on that screen is now everything it does.

You can select several projects in the studio and delete them together.

The Why this clip panel no longer repeats what was said. The same speech is already under the player, in sync with the video and clickable.

If YouTube turns a request away while your video is being fetched, the import keeps going: it switches connection on its own and picks up where it left off. It used to mark the video as failed and leave retrying to you — in our last run of ten videos back to back, four dropped this way.

When several videos are imported at once, the requests are spaced out instead of piling on top of each other, so bulk imports are turned away far less often.

Preparing a long video is measurably three times faster. Analysing the picture of a half-hour source took about twenty minutes; the same work now finishes in six or seven.

Two videos can be processed at once. The queue used to be strictly one at a time: with ten videos uploaded back to back, the tenth would not start until the first had finished.

The time-remaining estimate on the progress bar was retuned for the new speeds.

A faster-to-open format of the source video is preferred. Same resolution, same quality — it is simply much cheaper to prepare.

A clip is never shorter than the shortest length you asked for. A short burst of action could previously come back as a 10-second clip even when you had asked for 20 to 45 — the engine quietly halved your floor for anything that was not talking. It no longer does.

A moment that cannot fill the length you asked for is now left out instead of being stretched to reach it. Clips used to be padded with the seconds either side of the moment, which is footage that has nothing to do with what you clipped. The trade is deliberate: you may get fewer clips out of a video, but not shorter ones and not padded ones.

A clip's buttons sit in its title row now instead of on a separate strip underneath, and the advanced settings on the create screen fold down into the footer. Both give the clip itself more of the screen.

Joynubi now has a changelog. Every deployed version is listed newest first, with what changed in it, written in plain language.

The version you are running is shown in the footer of the home page and at the bottom of the sidebar in the app. If you report a problem, that is the number to quote.

The FAQ on the home page now answers what the product actually does today: which videos you can use, how long clips are, which aspect ratios you can export, how caption templates work, which features each plan includes, and what happens to the videos you upload.

A clip's status no longer stretches the clip list out of shape — it sits as a compact label instead.

Clips now use the whole length range you ask for instead of hugging the shortest allowed length. A clip keeps growing while the speech is still flowing, so the average clip came out at 37 seconds instead of 30 on our own test videos.

The create screen is one two-column form: link, clip length, clip count, language, caption template, video type, aspect ratio and layout are all visible at once. The preset strip and the project title box are gone — the presets only filled in fields you can now see, and an imported YouTube video overwrites the title anyway.

Select several clips and re-render them together. If one of them is refused — it is already rendering, or you are out of credits — the rest still go through and you are told which ones did not.

Sending a clip back to render without changing any setting now warns you that the result will be identical, instead of quietly producing the same file.

The rights-confirmation checkbox is gone from the import form; the responsibility note now sits in the footer next to the Terms of Service link. The aspect-ratio buttons carry the icons of the platforms each ratio suits.

Hook text — the line burned over the first seconds of a clip — is now a paid feature. Free accounts used to get it; from now on a clip made on a free account is rendered without it, and the field is locked with a link to pricing. Anything already written there is kept and comes back when you upgrade.

Hook text clears after the first three seconds of a clip. It was being drawn over the whole clip, so a line meant to catch someone in the first moments sat on top of the video until the end.

The clip page is now three columns: the clip list on the left, the player in the middle, the settings on the right. The player is roughly twice as tall and reels-shaped, its controls sit above the fold instead of below it, and it gained a volume slider and a playback-speed control — the subtitle indicator, which was only a picture and never a button, is gone.

A word-by-word transcript strip runs under the player. Clicking a word jumps the video to it.

Clip cards are horizontal with a vertical thumbnail, and the header above the list dropped from five rows to three, so about twice as many clips fit on screen. A clip that failed to render now carries a retry button instead of just a badge.

Select several clips and download them as a single archive.

Caption template previews are large enough to actually tell the templates apart, and one template can be applied to every clip in a project at once.

The per-clip caption editor — colour, outline, position — is gone. Captions now come from six ready-made templates instead.

Captions no longer sit across the speaker's mouth. The band was landing higher on the finished clip than it did in the preview — right over the face — and it now sits lower and in the same place in both.

Hook text fixes: a hook with no spaces in it is no longer drawn at all — clips were going out with a run of nonsense letters across the top, and the same string could be burned into a cover image. Suggested hooks are now written in the language spoken in the video rather than the language set for titles and descriptions.

Re-rendering a clip now actually shows you the new video. The player, the poster frame and the download link were all serving the previous render from cache.

Caption timing is right on videos that came from YouTube with hand-written subtitles. Those subtitles carry no real word timings — they were being spread across each line by character count — so the highlighted word ran up to a second behind what was being said. The spoken audio is now timed by us and the YouTube wording is fitted onto it.

The camera now cuts the clip at its own shot boundaries and plans each shot separately, picking one camera behaviour per shot instead of following the speaker frame by frame. Framing is calmer and subjects fall outside the frame far less often.

On the landing page, Start Free now puts the cursor in the URL box instead of scrolling a few pixels and doing nothing, on all three places it appears. The Studio opens with the create panel already open.