browser-use Releases Open Source AI Video Editing Skill video-use

browser-use released the open source Skill video-use for AI coding agents such as Codex and Claude Code, as announced on X.

The release centers on a pipeline that turns audio into compact text transcripts. ElevenLabs Scribe produces roughly 12 KB of output that includes word-level timestamps, speaker labels, and event markers. Agents invoke timeline_view.py only at decision points to create PNG frames.

This approach limits vision calls while keeping temporal data intact. The full pipeline runs transcription, packs data, builds a JSON EDL, renders with ffmpeg, and runs up to three self-evaluation rounds. Rendering steps include segmented extraction with copy codecs, 30 ms audio fades, PTS adjustments, final subtitle overlays, HDR mapping, vertical scaling, and two-pass loudnorm.

Animation engines such as HyperFrames, Remotion, and Manim are supported. Twelve hard rules are packaged with the project to enforce production correctness. The repository is available at https://github.com/browser-use/video-use and the announcement at https://x.com/shao__meng/status/2072644710523691110.

The change matters because most agent workflows still rely on heavy image sequences or full video context. video-use shifts the load to lightweight text while preserving edit decisions that can be rendered on demand.

Primary opponent in this story is the common practice of feeding agents dense frame sequences or long video clips. That route quickly hits context limits and cost ceilings. video-use replaces it with a timed transcript that agents query selectively.

Developers who already use Codex or Claude Code for scripting now gain a defined path for video tasks without rewriting their entire agent loop. The rules set reduces trial-and-error during render. For example, a developer could upload podcast audio, prompt the agent to isolate a speaker’s key segments via the timestamped JSON, then render a 60-second highlight clip using ffmpeg without manually scrubbing frames.

Questions remain about edge cases in speaker separation and complex motion graphics. The project does not claim exhaustive coverage for every production style. Early users may still need manual fixes when timestamps drift or animation engines produce mismatched frame rates.

Next signals to track are adoption logs from Codex users, updates to the twelve rules, and whether independent render tests confirm the claimed quality after three evaluation rounds.