Both hand the cut to an agent instead of a timeline, both run the edit on your own machine, and both are MIT. The difference is where the agent lives and what you have to assemble first. video-use is a skill you clone and symlink into Claude Code, with ffmpeg from Homebrew and an ElevenLabs key for transcription. VideoStudio is the video agent inside the Orkas desktop application, which brings its own ffmpeg and a local whisper runtime, so transcription needs no key and the audio stays on the machine.
Both are MIT, both edit real footage with an agent, and both render locally. One is a skill inside your coding assistant; the other is an agent in an application that already has the runtimes.
| Capability | Orkas VideoStudioThis site | video-useClaude Code skill for video editing |
|---|---|---|
| Editing from inside the assistant you already run | Lives in a separate desktop application, which hosts Claude Code rather than living inside it | Symlinks into Claude Code's skills directory |
| Community behind it | A much smaller project | About 25,100 GitHub stars, from the browser-use team |
| Animation overlay backends | HTML composition rendered frame by frame | HyperFrames, Remotion, Manim or PIL |
| Transcribing the footage | Bundled whisper.cpp runs locally — no key, audio stays on the machine | Requires an ElevenLabs API key |
| What you install first | The application; ffmpeg and the transcription runtime ship inside it | Clone, symlink, uv sync or pip install -e ., ffmpeg via Homebrew, then a .env |
| Work that is not the video | Sibling agents in the same app cover documents, slides, images, research and code | Video editing is the scope |
| Stopping before it changes the cut | An edit decision list held at an approval gate; changing cuts or delivery invalidates the signed plan | Ask, confirm, execute, self-evaluate, persist — never touches the cut without strategy approval |
| Licence | MIT | MIT |
The first three rows are hard limits here, not preferences. VideoStudio ships inside a separate application and cannot be symlinked into a coding assistant you already run; its community is a fraction of the size; and its overlays come from one HTML compositor rather than a choice of four backends. If any of those is what you came for, video-use is the answer and nothing below changes it. Note also what ties: both stop and ask before they cut, and both are MIT.
This comparison was last updated on September 18, 2026; check each vendor's site for current terms.
You probably should not pick one — Use each for the job it is built for, and pass the resulting files into Orkas when the deliverable needs stills, slides or a written script alongside the video.
Keeping both costs nothing — both are MIT and both work on the same folders.
Leave the cuts you drive from inside Claude Code with video-use.
Send the ones that also need a deck, stills or a written script to Orkas.
Both read and write ordinary files, so nothing has to be migrated between them.
Both are MIT licensed, both run the edit on your own machine, and both drive the cut with an agent rather than a timeline. The difference is shape. video-use installs into Claude Code as a skill, so the agent you already run does the work. Orkas VideoStudio is the agent on the other side, and it lives in a separate desktop application that also carries review gates and other production lines. If you want the editing to happen inside the assistant you already use, video-use is the answer and nothing here changes that.
No. VideoStudio is an agent inside a desktop application; it cannot be symlinked into another agent's skills directory the way video-use is. Orkas can run Claude Code as a local session from inside itself, which is the reverse arrangement: the app hosts the CLI rather than the CLI hosting the editor. If your requirement is that nothing new gets installed beyond a skill folder, that requirement rules Orkas out.
video-use lists an ElevenLabs API key as a requirement for transcription, so the audio leaves your machine and the usage is billed per call. The Orkas app bundles a whisper.cpp runtime and its own ffmpeg build, so VideoStudio transcribes locally with no key and no per-minute charge. This is about the transcription step only — generated shots on either side still use whichever video model you point them at, and those are billed by that provider.
video-use documents cloning the repository, symlinking it into the agent's skills directory, installing dependencies with uv sync or pip install -e ., installing ffmpeg through Homebrew, and copying .env.example to .env. Orkas is an application install; ffmpeg and the transcription runtime ship inside it, and VideoStudio is already on the team. Neither approach is wrong — one assumes a Python toolchain is already part of your setup, the other assumes it is not.
Yes, and this is a genuine tie. video-use describes its loop as ask, confirm, execute, self-evaluate, persist, and states that it never touches the cut without strategy approval. VideoStudio puts an edit decision list in front of you, holds it at an approval gate, and treats a change to the cuts, copy, source allocation or delivery settings as invalidating the signed plan. Both refuse to edit silently; neither is meaningfully safer than the other on this point.
video-use, clearly. Its repository carries roughly 25,100 stars at the time of writing, and it comes out of the browser-use team. Orkas is a much smaller project. Stars are not a measure of whether a tool fits your job, but if what you want is a large user base, prior art in issues, and many people to ask, that argument favours video-use.
Free, MIT licensed, with ffmpeg and local transcription already inside.
