Launched this week
Local Video Studio
Make AI videos locally, one restartable stage at a time
2 followers
Make AI videos locally, one restartable stage at a time
2 followers
Local Video Studio turns a topic into a structured video project using your own models and hardware. Plan scripts, review scenes, generate voice and visuals, align captions, inspect the timeline, and render a final MP4. Every stage is saved and can be rerun independently, while prompts, seeds, attempts, media, and renders stay in portable folders. It is open source (Apache 2.0), local-first, has no telemetry, and includes a no-GPU mock mode. Gemini TTS is optional and opt-in.
I built Local Video Studio because my local AI video workflow had become a pile of model UIs, scripts, temporary files, and expensive restarts.
The core idea is simple: video generation should behave like a restartable production pipeline, not one giant prompt. If one scene is wrong, regenerate that scene. If the narration changes, rebuild only the affected downstream stages. Keep the prompts, seeds, model versions, attempts, approvals, and final assets in a project folder you can inspect and move.
The studio currently brings together local planning, still images, audiovisual shots, voice cloning, music, caption alignment, exact-text graphics, thumbnails, and deterministic FFmpeg rendering. It also includes a complete mock mode so anyone can test the workflow without downloading model weights.
I would especially value feedback on three things:
Which local model backend should be easier to set up next?
Where does the first-run experience still feel too technical?
Which part of a real video workflow is missing from the current pipeline?
The source and installation guide are on GitHub. If the local-first approach is useful to you, a star or a specific issue would help the project reach more builders.