simbioclip

the problem

Repurposing long-form video into shorts means paying a monthly subscription per seat, and handing raw footage to a third party you do not control. For anyone publishing daily, the cost scales with output and the footage never stays private.

what i built

A self-hosted pipeline that runs the whole job end to end: pull the source with yt-dlp or a direct upload, transcribe with word-level timestamps, let an LLM pick the moments worth cutting, then render vertical clips with burned-in captions in FFmpeg. FastAPI takes the request and queues it in Redis; RQ workers run the stages independently so a long render never blocks the next download. The dashboard is HTMX, so job status updates live without a frontend build step. Supports 9:16, 1:1, 4:5 and 16:9 output with center-crop, talking-head and blurred-background layouts, plus per-frame trimming on individual clips. Ships as Docker Compose and runs on a plain VPS. Speech-to-text can run locally with faster-whisper or hit a remote endpoint, and the LLM layer is OpenAI-compatible so it works with Groq, OpenRouter or Google AI Studio.

stack