// Agent profile
browser-use/video-use is an open-source agent-driven video editor from the team behind browser-use (the most-starred browser automation agent on GitHub). The workflow is conversational: drop raw footage in a folder, tell your coding agent what you want, and get final.mp4 back. The agent reads videos through audio transcripts and visual analysis rather than watching them, cuts filler words and dead space, auto color grades, burns styled subtitles, and generates animation overlays via HyperFrames, Remotion, Manim, or PIL — spawning parallel sub-agents for each animation. At 23,000+ GitHub stars, video-use has become the most popular agent-to-video tool on GitHub. It works with Claude Code, Codex, Hermes, or any agent with shell access, and persists session memory so editing projects carry context across sessions. Self-evaluation at every cut boundary catches errors before showing output.
video-use is an agent-driven video editor from the browser-use team, currently at 23,000+ GitHub stars. The premise is that video editing is a sequence of decisions that a coding agent can make by reading transcripts and visual metadata rather than watching footage in real time. You drop raw takes in a folder, chat with your agent, and get a finished video back — no timeline UI, no presets, no menus.
The technical approach avoids giving the LLM raw video frames. Instead, video-use builds two analysis layers. Layer one is an audio transcript via ElevenLabs Scribe — word-level timestamps, speaker diarization, and audio events like laughter or applause, packed into a compact text representation the agent can reason over. Layer two adds visual analysis for shot composition, motion, and scene boundaries. Together, these give the agent enough information to make word-boundary-precise editing decisions.
The editing pipeline handles the full post-production stack: filler word removal, dead space trimming, color grading (warm cinematic, neutral punch, or custom ffmpeg filter chains), 30ms audio fades at every cut to eliminate pops, and styled subtitle burning. Animation overlays are generated in parallel — each one spawns its own sub-agent using HyperFrames, Remotion, Manim, or PIL depending on the animation type.
The self-evaluation loop is a key design choice. After rendering each segment, video-use checks the output at every cut boundary before presenting it. This catches misaligned cuts, audio artifacts, and visual glitches before the human ever sees them.
Open-source game development framework powered by Claude Code — autonomous agent loop for designing, iterating, and playtesting games using natural language.
OpenAI's AI image generation model that creates and edits realistic images from natural language descriptions.
AI voice platform that generates ultra-realistic speech, voice cloning, and audio content from text with human-like expressiveness.