sitor-ai/frametell

Frametell – AI‑powered explainer‑video creator

What it is

• A desktop application (macOS Apple‑Silicon) and CLI tool that turns a short idea, notes, PDFs, slides, or images into a 1080p narrated video.
• Uses a director agent (LLM) to draft a storyboard, write a script, generate voice‑over, draw each page in a chosen visual style, and finally render the film.
• All processing runs locally; only the LLM and TTS services are called via API keys you provide.

Why it matters

• Lets knowledge creators produce polished explainer videos without learning video‑editing software.
• Gives you a chance to review and edit the storyboard before any rendering occurs, reducing wasted API tokens.
• Works with many LLM providers (Anthropic, OpenAI, Google Gemini, DeepSeek, OpenRouter, etc.) and with ElevenLabs or Volcengine TTS for voice‑over.

Key features (as described in the README)

• Storyboard creation – LLM turns your input into a sequence of pages and story beats; you can approve or edit them.
• Narration & voice – Automatic script writing and optional voice synthesis (ElevenLabs for English, Volcengine for Chinese).
• Page drawing – Generates each frame in one of several built‑in styles (warm‑story, study, chalkboard, whiteboard, tech‑dark).
• Quality checks – Linting and screenshot previews for each page; final video check before rendering.
• Local execution – Your data and API keys never leave your machine; the heavy lifting (video rendering) is done with a bundled FFmpeg and a headless Chromium/Edge instance.
• Cross‑model support – Plug‑in any OpenAI‑ or Anthropic‑compatible endpoint.
• Export – 1080p MP4 with subtitles and publishing notes.

How it works (step‑by‑step)

1. Provide input – type a sentence, paste notes, or drop a document (PDF, Markdown, slide deck, images).
2. Storyboard generation – the LLM (referred to as the director) proposes a sequence of pages; you approve or modify.
3. Script & voice – the agent writes the narration and calls ElevenLabs (or Volcengine) to synthesize speech.
4. Page rendering – each storyboard page is drawn in the selected style; the system runs lint checks and shows screenshots.
5. Film assembly – a final consistency check is performed, then FFmpeg stitches the frames, adds subtitles, and outputs a 1080p MP4.

Getting started

• macOS (Apple Silicon) – download the signed .dmg from the latest release, install, then add your LLM and ElevenLabs API keys in Settings.
• From source – clone the repo, install Node 22+, pnpm, and a Chromium‑based browser. Run pnpm install && pnpm build, set the required environment variables (ANTHROPIC_API_KEY, ELEVENLABS_API_KEY, etc.), and start the web UI with node apps/cli/dist/cli.js serve or create a video directly via node apps/cli/dist/cli.js run notes.pdf --style warm-story --duration 60 --lang en.

Community & support

• Discord community for questions, sharing videos, and feature ideas.
• Issues can be opened on GitHub for bugs or feature requests.

License

• AGPL‑3.0 – free to use, modify, and run, but network‑distributed modifications must also be open‑sourced. A commercial license is available on request.

---

Frametell is a genuine AI‑driven tool that automates the end‑to‑end pipeline of creating short explainer videos, positioning it squarely within the frontier of generative AI and multimodal content creation.