Keep the motion. Change the character.
A local, open-source video studio with your original soundtrack intact.
Download · Quick start · Providers · Feedback
| Motion guidance | Character picker | Original audio |
|---|---|---|
| Prepare colored depth and subject masks, then review the whole clip. | Upload, replace, remove, or reuse references saved in local jobs. | Export with the complete source soundtrack restored. |
Choose a supported model from a searchable picker. Review the exact request and available price estimate before generating. Follow progress and return to saved results in History.
The app is free and MIT licensed. Hosted preparation and generation use your own paid provider accounts.
Download the latest source ZIP and extract it, or clone:
git clone https://github.com/abdullayevziyodulla/open-genjutsu.git
cd open-genjutsuInstall Python 3.12+ and Rubber Band CLI. On Windows, put rubberband.exe on PATH or in tools/rubberband. On macOS use brew install rubberband; on Ubuntu use sudo apt install rubberband-cli python3-venv.
Windows
python -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txt
.venv\Scripts\python.exe start-windows.pyAfter installation, you can double-click Start-Open-Genjutsu.cmd. Use Stop-Open-Genjutsu.cmd to stop the studio.
macOS / Linux
python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt
.venv/bin/python launch.pyUse PORT=8771 .venv/bin/python launch.py to select another port.
Open http://127.0.0.1:8770. The Windows launcher picks another available port if necessary.
Open API settings in the studio, or copy .env.example to .env.
- Replicate token: required for vocal separation, depth and subject masks.
- Video provider: choose Oxen, Replicate or Lec and add your own key.
- Oxen storage: use an existing private
namespace/repositoryfor generation inputs.
Keys stay in your local .env. Selecting files and models does not submit a generation. Inputs go to the selected providers when you explicitly prepare or generate.
01 · Add motion — Select the original video. Prepare its guidance with your Replicate account. Preparation is billed.
02 · Add identity — Open the character picker. Upload a portrait or supported identity video, or reuse a local reference. A second slot means a different person, not another angle of the first person. Wan accepts one photo.
03 · Review & generate — Watch the entire prepared clip and approve its mask or tracking. Choose your model, review the exact request and price estimate, then generate.
04 · Keep the soundtrack — Download the finished video in Results. Supported HD renders require another deliberate action and separate billing.
| Model / provider | Starting output | Higher resolution |
|---|---|---|
| Seedance 2.5 · Oxen | 480p draft | 1080p render of the saved draft |
| Seedance 2.5 · Replicate | 480p preview | New 720p generation |
| Wan 2.2 Animate · Replicate | 480p preview | New 720p generation |
| Seedance 2.5 · Lec | 720p | No separate HD step |
The quality picker shows what each adapter actually supports. Higher resolution is offered after a draft is ready. Availability, moderation, pricing, and identity or motion fidelity depend on the provider.
What happens during preparation?
Default preparation isolates vocals, pitches them +3 semitones without changing timing, and embeds them in a colored-depth composite with the subject mask. The source remains intact. The final export restores the complete untouched original soundtrack.
Optional local MediaPipe guidance can be installed with bash setup-face-mesh.sh using Python 3.12. Mesh modes are experimental and work best with one centered speaker. Wan uses the original video with its own tracking rather than the depth composite.
Bug reports and contributions are welcome. Start with CONTRIBUTING.md. Report issues with sanitized errors and reproduction steps; keep keys and private media out of public reports.
# On Windows, use .venv\Scripts\python.exe instead.
.venv/bin/python -m unittest discover -s tests
.venv/bin/python test_composite.py
.venv/bin/python scripts/package_release.pyThe offline checks run in GitHub Actions. The release exporter includes source and project artwork; it excludes credentials, saved jobs, private media, environments, models and Git history.
Setup · Workflow · Troubleshooting · Security
This is a localhost, single-user studio. Public multi-user hosting needs authentication, isolated storage and a durable worker queue.
Open source. Bring your own keys. Make it yours.
MIT license · Required attribution · Third-party terms. Independent project; no affiliation with a commercial video platform. No model weights or third-party demonstration footage are bundled.
Open Settings → Language to choose English or O‘zbekcha. English is the default. The choice is saved in your browser and also applies to the setup guide. Prompts, API keys, filenames, and provider responses stay unchanged.