Beta · changelog
What's new in the studio
Every beta release, newest first. The latest is beta.16. In the studio, the New version ready card in the left menu shows what you don't have yet and installs it.
On beta.11 or older, run npm run update once first.
- The launcher starts again. Beta.15 had a mistake in its start script, so the launcher built the app and then stopped with
SyntaxError: Unexpected reserved word. If that happened to you, open a terminal in the studio's folder and run npm run update. It needs your license key. Then start the launcher again.
- Don't press Update while running
npm run dev. That stops the update halfway. Stop the studio and run npm run update in a terminal instead. If it already happened and the build fails, run npm run update -- --force. From this version the button says so itself.
- Your login writes your scripts. Logging in with ChatGPT or Claude in Settings now picks that login for scripts, unless you already chose a provider that works. Script model lists your logins next to your keys and shows what scripts really run on.
- Pick a model for Codex. Script model lists the models your Codex offers. If your plan or Codex version can't run one, the script stops and says so.
npm run setup:comfyui checks the models too. It asks the running ComfyUI for every file before the test render. If another ComfyUI install holds the port, it stops and tells you what to do.- Bring your own script. My script in the draft panel uses your words as written. An empty or numbered line starts a scene, you see the scenes and the rough length, and no AI runs.
- Pictures from another tool. Pictures for all scenes gives you prompts to copy and a tile per scene to drop a picture on. Drag a picture onto another scene to swap them.
- A simpler dock. Each pane shows its small settings first, and anything bigger opens as a row with a way back. Generating a picture has its own window, with a Cast tab. The rail names its tools.
- A new New channel window. Name, Colours and Look are three steps in one window. Colours can come from a website or pasted codes, and you edit the look on the reel itself.
- Back to the pitches. You can reopen the 3 pitches from the draft panel and, after a pick, from Edit › Pitches in the Script panel. They are saved with the reel.
- Studio palettes. A swatch under the logo picks Broadcast, Graphite, Dusk or Moss. It changes the studio's panels, never a reel.
- Caption styles play a small sample of themselves, and image prompts leave room for captions.
- A pitch from an article now speaks as its author. Windows sit in the middle of the screen. A new reel no longer says it differs from its channel.
- Your whole script as text. Edit › Edit/Export in the Script panel shows it as Markdown. Copy it, download it as a .md file, or change it and press Apply, which is one undo step. Each scene keeps its pictures, timing and voice as long as its heading keeps its number.
- A calmer Script panel. Dramaturg, Draft with AI and Edit/Export sit under one Edit button.
- Remove scene shows the scene first in the Dramaturg, so you can read what you would cut.
- Ctrl on Windows and Linux. Shortcut labels now show your own keys.
- Scripts stay in your brief's language. One that switches gets another try, then stops with a message.
- Recent errors in Health no longer lists Node's own notices or routine script repairs.
- A browser extension can't raise a hydration error on
npm run dev any more.
- An update stopped halfway can now be run again. It used to say "Already up to date".
- Motion Designer suggestions sit centred, and a checklist holds only items.
- The Update button finishes the job. An update to beta.14 could stop at "The new version could not be prepared" and leave the studio unable to build. Updating now keeps everything the build needs. The studio still builds after an update, which takes a minute or two.
- If your studio stopped there, open a terminal in its folder, run
npm install --include=dev, then start the launcher again.
- Windows fixes, and a short build after this update. Downloads since beta.10 carried an app built on a Mac, and that removed code Windows needs: Codex read as not found even when installed, starting ComfyUI looked for the wrong Python file, stopping a render used the Mac way of ending it, and the publishing tunnel fetched the Mac version of its helper. The studio now builds on your own computer at first start and after each update. It takes a minute or two and the black window shows the steps, so keep it open.
- Cast library. Each person or product in a channel's cast gets a gallery of up to 24 reference pictures, a role per picture and an identity note. A scene picture sends only the references that fit its shot, and Sends N references shows which. Codex, OpenAI and Gemini all take them.
- Motion and Background for components. Plain components get a Motion section for how they enter, move and leave. One Background control with previewed tiles replaces the Variant select.
- Publishing keeps every YouTube, Instagram and Substack account in one list. Substack gets named accounts, and account rows can be edited.
- Karaoke word colour now reaches the stage and the take. A reel whose Karaoke word colour differs from its accent changes on its next render.
- Pitches stay in your brief's language. A card that switches language goes back to the writer once.
- GPT-5 works as the script model.
- A missing Codex now says it is not installed and where to log in. Missing ComfyUI models fail at once with the full list, without two more attempts.
- The Voice tab says when a reel still speaks with another provider and offers the switch. During Play reel, the music waits with the picture while a later scene's voice loads.
- Instagram calls retry after a network failure, and Settings › Accounts & keys no longer jumps when you switch tiles.
- Music that runs out. When the reel's music track is shorter than the reel, the Music tab says where it stops and offers Loop to fill, Stretch to fit (up to 15 % slower, same pitch) or Extend the track – ElevenLabs continues the same music to the reel's length.
- Music that stops early. A music bed that no longer reaches every scene (you added scenes, or moved its last one up) names the silent scenes, and Cover the whole reel stretches it over all of them.
- Generate in the Music tab makes at least a minute of music, so a reel that grows later still has music under it.
- Drag the playhead. While Play runs, drag the line on the filmstrip – or press and drag on any scene – to hear a part again or skip ahead.
- Seek in a take. Clicking the timeline of a finished take now jumps there.
- Play shows transitions. Play reel plays each scene's tail, so the transition into the next scene shows in the preview.
- Animation time estimates follow your GPU profile (Eco, Balanced, Max) and the number of Long shot windows; while Render take animates, it counts down the time left.
- Watermark (Brand): a faint corner image on every branded scene, draggable on the stage.
- Captions can be switched off on a single scene; Generate keeps the prompt you typed for each scene.
- Component lists are edited one row per item – clearing a point no longer deletes the whole list.
- Picking an ElevenLabs narrator in the Voice tab switches the reel to ElevenLabs.
- Assistant and Feedback sit together in one small pill bottom-right.
- Update from the studio. A New version ready card in the left menu shows what's new in every version since yours, and Update installs it and restarts the studio – a running render finishes first.
npm run update knows that beta.12 is newer than beta.9 (it compared beta versions wrongly) and puts the new build in place last.- Coming from beta.11 or older: this one update still goes through
npm run update, then start the launcher again. From here on the card does it.
- Scripts in your article's language. A Portuguese (or any other) article now gets a script, cards, captions and a call to action in that language – before, every script came out in English.
- Captions cut like subtitles. A caption page never ends on the first word of the next sentence or on a small word like "e", "em" or "the"; long sentences split into even pages, preferably at a comma, and no single word hangs on its own line.
- Pages & line breaks (Captions tab). Set caption line and page breaks by hand, join pages, and rewrite what a page shows (say "34%" where the voice says "thirty-four percent").
- Find a voice (Voice tab). Search the ElevenLabs Voice Library by language, accent, gender and use, play free samples, and use a voice in one click.
- Add a voice by ID or link and ↻ Refresh in the Voice tab – a voice you added on elevenlabs.io shows up without restarting.
- ElevenLabs key check (Settings › ElevenLabs): what your key can do, permission by permission, and a link to create a key.
- ElevenLabs problems say what to fix: a missing key permission, an account out of credits, or a voice no longer in your account now each get their own message instead of "rejected the API key".
- The download ships a prebuilt app, so the first start skips the build.
npm run update replaces the old build instead of serving it.
- The "Connect your AI" banner explains what each AI covers: Claude alone writes scripts; scene images need ChatGPT, OpenAI or Gemini.
- Feedback button for beta testers.
- The assistant saves the open reel before it changes anything.
- With Kokoro, captions follow the voice: words are placed on the measured speech, sentence by sentence.
- Log in with Claude or ChatGPT in Settings – no terminal needed.
- The first render works on a clean Mac (ffmpeg for the voice is bundled).
- Settings are organised in tabs.
- The launchers install Node.js themselves (one system prompt).
- A quieter install, and help for the Windows and macOS prompts in the README.
- Help for the macOS Gatekeeper warning in the README.
- The launchers say honestly how long the first install takes.
- Fixes the Windows build that failed on a fresh machine in beta.2.