- Category
- AI Agents
- Rank
- No. 820Tools index
Previous survey · No. 815 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- browser-use
- GitHub
- 24.4k stars
- Date
About
Edit videos with coding agents — Browser Use's experiment: describe the edit in plain English and let an agent execute it across the timeline.
What it does
Video Use turns footage into a transcript-led editing project. It transcribes each source with word timestamps, speaker labels, and audio events, then condenses the results for an agent to reason over. Filmstrip and waveform composites provide visual checks only when needed. The agent builds an edit decision list, renders segments with FFmpeg, adds grading, overlays, and subtitles, then inspects cut boundaries before presenting a preview.
Why it's ranked here
The architecture is unusually concrete for an agent experiment. Transcript caching, word-boundary cuts, audio fades, output-timed captions, and post-render checks address failures that often make automated edits unusable. The trade is substantial setup and external transcription dependence, while artistic decisions still rest on the agent following a detailed operating guide.
What's good
Its strongest idea is selective perception. The agent reads compact, timestamped speech instead of consuming every frame, then requests combined filmstrip, waveform, word, and silence views around uncertain moments. Rendering also encodes practical safeguards: per-segment processing, lossless concatenation, caption placement after overlays, HDR tone mapping, portrait-aware scaling, and bounded automatic color correction.
Tradeoffs
Transcription requires an ElevenLabs API key and uploads extracted audio to its speech service. FFmpeg and FFprobe are mandatory, while some animation paths add Node.js or optional creative tools. Installation registers the repository as an agent skill rather than providing a polished standalone application. Quality also depends on the agent respecting a long process document, and self-evaluation stops after three passes even if problems remain.
How to use it well
Use it for speech-led projects with multiple takes, such as launch videos, tutorials, interviews, and talking-head edits. Let it inventory and transcribe footage, review its proposed structure, then approve rendering and iterate through natural-language feedback. It fits an agent-assisted production workflow where reproducible cut mechanics matter. It does not replace hosted transcription credentials, local media tooling, or a conventional visual timeline editor for hands-on frame-by-frame control.
Technical notes+
pyproject.toml defines a Python 3.10+ setuptools project with requests, librosa, matplotlib, Pillow, and NumPy, plus optional Manim support, but declares no console scripts. helpers/transcribe.py extracts mono 16 kHz WAV audio with FFmpeg and calls ElevenLabs Scribe for diarization, audio events, and word timestamps. helpers/transcribe_batch.py adds cached parallel processing. helpers/pack_transcripts.py groups words on silence or speaker changes. helpers/timeline_view.py builds PNG filmstrip and waveform composites. helpers/render.py consumes an EDL, extracts and grades segments, concatenates them, handles HDR sources, and applies subtitles after overlays. helpers/grade.py provides bounded automatic corrections, presets, and raw FFmpeg filters. SKILL.md supplies the agent workflow and production rules; install.md covers skill registration and system dependencies.
Observed
- License
- MIT License
- Primary language
- Python
- Python requirement
- Python 3.10 or newer
- Packaging
- Setuptools project installable with uv sync or pip editable install; no console scripts declared
- Interface
- Agent skill plus directly invoked Python command-line helpers
- Required services
- ElevenLabs Scribe API for transcription
- Required media tools
- FFmpeg and FFprobe
- Optional tooling
- yt-dlp for online sources; Manim, HyperFrames, or Remotion for animation work
Read from README.md, pyproject.toml, helpers/grade.py, helpers/render.py, helpers/transcribe.py, helpers/timeline_view.py, helpers/pack_transcripts.py, helpers/transcribe_batch.py, LICENSE, SKILL.md, install.md, poster.html.
What it can do
Execute video editing commands from natural language descriptions
Plain English editing instructions → Executed video edits on timeline
Automate timeline-based video editing tasks
Video file and editing description → Modified video with applied edits
Interpret natural language for video manipulation
Text descriptions of desired video changes → Corresponding video editing operations
Apply coding agent logic to video editing workflow
Video content and user instructions → Programmatically edited video sequence
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.