VideoSOS: 100+ AI Video Models in Your Browser
Most AI video tools are either a closed SaaS with a monthly bill or a research demo with no editing workflow. VideoSOS tries to be the third option: an open-source, browser-based video editor that plugs into 100+ commercial AI models — Veo 3.1, Kling, FLUX, Imagen — through fal.ai and Runware, with a real timeline, cost tracking, and nothing uploaded to anyone's server. At just 1,265 stars, it is the hidden gem of this roundup.
What it does
VideoSOS is a Next.js application that combines two things that rarely ship together: access to 100+ AI generation models (text-to-video, image-to-video, text-to-image, image editing, music, and voiceover) and a real multi-track timeline editor with drag-and-drop, keyframes, audio waveforms, and 16:9 / 9:16 / 1:1 aspect ratios for social platforms. Generation runs through your own fal.ai and Runware.ai API keys — you can use either provider or both — and the README documents the model catalog in unusual detail, with a model-update script (npm run update-fal-models) that syncs models and pricing from the provider APIs.
The architecture is the selling point for privacy-minded creators: everything runs locally in your browser. Projects, media, and API keys live in IndexedDB on your device — no uploads, no tracking, no data collection. Rendering is client-side too, via FFmpeg.wasm and Remotion. There is even a live demo you can try before installing anything.
Two features show the author actually makes videos: built-in cost tracking (per-item generation costs, project totals, breakdowns by provider and model — essential when every clip is metered) and real-time pricing display for FAL models before you hit generate.
Who should use it
- Budget-conscious AI video creators who want Veo/Kling/FLUX access without another SaaS seat — you pay the providers directly, per generation.
- Privacy-first producers — client projects and unreleased concepts never leave your browser.
- Experimenters comparing models side by side — the dual-provider catalog and pricing display are built for exactly this.
- Self-hosters — Docker support and a portable build are documented.
Getting started
The fastest path is the live demo — no install. To run it yourself:
git clone https://github.com/timoncool/videosos cd videosos npm install npm run dev
Then open http://localhost:3000 and add your fal.ai and/or Runware API keys in the Settings dialog (gear icon). Keys stay in your browser's local storage. Docker users can docker compose up -d instead.
Honest pros and cons
✅ Strengths
- 100+ models behind one timeline — unusually broad catalog
- Genuinely private: browser-local storage, client-side rendering
- Cost tracking and per-model pricing before generation
- MIT licensed, live demo, Docker and portable builds
- Automated scripts keep the model/pricing catalog fresh
⚠️ Tradeoffs
- The software is free; the generations are not. Every AI call bills your fal/Runware balance — budget accordingly
- Small community (1.2k stars) and essentially a solo-maintainer project — bus-factor risk
- Model availability follows the providers; catalog entries can drift between updates
- Browser-based rendering is convenient but slower than native for heavy timelines
- English/Russian i18n only, at the time of checking
Verdict
VideoSOS is the most practical "try 100 models without 10 subscriptions" setup we have found in open source. It will not replace a full NLE for complex edits, and it does not make the AI calls themselves free — but as a private, cost-transparent workbench for AI-assisted video, it punches far above its star count. If you generate a lot of AI clips and currently juggle provider playgrounds in browser tabs, this consolidates the mess into one timeline.