Image: GitHub
Why it mattersThe storyboard approval step gives the user a decision point before any page is drawn. That lowers the cost of a wrong direction from a whole render to a few seconds of reading.
Making an explainer video with AI is still a cost gamble. You feed a script in, pay for a render, and find out at the end whether the pacing works. Frametell is a native Mac app that moves the decision point earlier. A director agent reads your idea, writes a storyboard, and asks you to sign it off before anything is drawn. The repository has 402 stars. Sitor AI created it on 10 October 2026, 16 days ago.
What the agent does
One sentence is enough to start, such as "explain the uncertainty principle". You can also hand it notes, a PDF, a slide deck or images. The director writes pages and story beats into a storyboard. The user approves or edits the storyboard. The agent then writes the narration and voices it with ElevenLabs, or Volcengine TTS for Chinese. It draws every page in the chosen style. It checks each page with lint and a screenshot. A final run on the full video renders a 1080p MP4 with subtitles.
One demo, with its numbers
The README publishes one film with its real measurements. The film is 72 seconds long. The maintainers say the run took 12 minutes 52 seconds and cost $1.59 of model time. That is one film, from the project's own test. Frametell has five built-in styles: warm-story, study, chalkboard, whiteboard and tech-dark. Two demo films linked from the README cover Rayleigh scattering and compound interest. Both used Gemini as the model.
How to run it
The macOS app is on Apple Silicon only. The maintainers say the app is signed and notarized by Apple. The app also needs Chrome, Edge or Chromium installed, because Frametell uses it to draw and render the pages. Windows and Intel Mac users must run it from source. The source build needs Node.js 22.19 or later and pnpm. FFmpeg is bundled. The user supplies model keys for a provider such as Anthropic, OpenAI, Google, DeepSeek or OpenRouter. The ElevenLabs key for the voiceover is optional.
The licence
Frametell is released under AGPL-3.0. The maintainers write that a user may run and modify it, including for a paid job. The condition is that a modified version offered over a network must be released under the same licence. The project offers a commercial licence without that obligation on request. The maintainers call Chrome a drawing engine and say it does the rendering, not a model.
A team used to pay for an unseen video. The agent reads a script, picks visuals, and renders a file. The team finds out whether the pacing works after the invoice. Frametell changes the order. The storyboard is the decision point. A page the user does not want is caught before it is drawn, voiced or rendered. The pattern is small, and it moves a cost. One wrong direction costs a few seconds of reading, not a whole render.
Source
- Frametell: sitor-ai/frametell on GitHub
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.