Open Genjutsu · openjutsu in the browser
Open Genjutsu — Swap the Person in Your Video Without a GPU
Open Genjutsu by TryH3 works like openjutsu in your browser: your photo replaces your clip’s person, the model is told to follow its moves and camera, and its sound returns.
Seedance 2.5 · 5 to 30 seconds · from 55 credits · no watermark · nothing to install
Open Genjutsu on TryH3 is for people who found openjutsu or open-genjutsu on GitHub but have no NVIDIA GPU, no ComfyUI and no API key. Upload a 5-second to 5-minute clip or pick a built-in one, add one or two photos, and get a 5 to 30 second MP4 at 480p, 720p or 1080p with the original sound and no watermark. Videos use purchased credits: from 55 (about $1.52); the default 10-second 480p video is 108.
Updated 2026-11-19
How Open Genjutsu works in the browser
Open Genjutsu takes four inputs on one screen, with nothing to install, and nothing is charged until you press the button.
- 01
Pick a clip
Upload an MP4 or MOV of 5 seconds to 5 minutes (up to 50 MB), or choose one of 24 built-in people clips. Then choose which 5, 10, 15 or 30 seconds to use.
- 02
Add the photo
Add one photo of the person or character who should appear, or two for a two-person clip. Image 1 replaces the most prominent person, Image 2 the next one; a short note can change who replaces whom.
- 03
Choose length and resolution
Open Genjutsu here runs on Seedance 2.5, which edits your clip directly rather than drawing a new one. 480p is the default; 720p and 1080p cost more. The exact credit cost shows before you generate.
- 04
Generate and download
The video renders in a few minutes. The original sound of the selected seconds is laid back under it, and the MP4 downloads without a watermark. If the video fails or is rejected before it is made, its credits return.
Open Genjutsu examples
Each clip plays the original video and the swapped result, most with the photo in the corner, so you can compare the motion, the background and the new person.
Open Genjutsu GitHub, app or download: which way should you run it?
There is no single Open Genjutsu download: the name covers several GitHub projects. D3voz/openjutsu runs the whole swap on your own NVIDIA GPU through ComfyUI. Anil-matcha/open-genjutsu and abdullayevziyodulla/open-genjutsu are local apps that send each job to a paid video API with your own key. This page is the no-install route: the same kind of swap in a phone or desktop browser, paid per video in credits. Pick by what you already have: a strong GPU, an API account, or neither.
| Route | What you install | Hardware | Who bills you | Clip limits | Original sound |
|---|---|---|---|---|---|
| D3voz/openjutsu (local ComfyUI) | ComfyUI, its custom node packs and model files, then the Openjutsu app | An NVIDIA GPU; tested on an RTX 4060 Ti 16 GB with 64 GB of system RAM | No per-video fee; your own hardware and electricity | Uploads up to 512 MB per request; a 6 s 480p clip took about 6–13 minutes on the test card | Kept from the source video |
| Anil-matcha/open-genjutsu (local app + Muapi) | Python 3.10+, FFmpeg and a Muapi API key | Any computer; rendering runs on Muapi | Your Muapi account, per image, analysis and video stage | Source 4–15 s, up to 50 MB; up to 8 character images; 4–30 s output | Restored on export |
| abdullayevziyodulla/open-genjutsu (local app + your keys) | A Python app, a Replicate token and an Oxen, Replicate or Lec key | Any computer; rendering runs on the provider | Your provider accounts; preparation is billed too | Seedance 2.5 or Wan 2.2 Animate; preview first, then a 720p or 1080p render depending on the provider | Full original soundtrack restored |
| Open Genjutsu on TryH3 (this page) | Nothing; a phone or desktop browser | None | Credits per video, from 55 (about $1.52) | Clip 5 s to 5 min, up to 50 MB; 5–30 s output; up to 2 photos | Audio of the selected seconds added back |
Checked 2026-11-19 against each project’s README. The open-source apps are MIT licensed; model licences are separate, and model names, limits and fees can change with each release.
Is open source Genjutsu free?
The open-source Genjutsu apps are free to download, but the videos are only free on your own hardware. D3voz/openjutsu has no per-video fee if you own a capable NVIDIA GPU; the two open-genjutsu apps send every job to a paid API on your own account. On this page every video uses credits: the cheapest is 55 credits (about $1.52), the $5.00 Trial pack of 24 credits does not cover one, and the smallest pack that covers the default video is $9.90 for 180 credits.
| Sign-up credits | Do not unlock Seedance 2.5; a credit purchase does |
|---|---|
| Trial pack | $5.00 for 24 credits: below the 55-credit minimum |
| Smallest pack that covers the default video | $9.90 for 180 credits (one-time, credits do not expire) |
| Watermark | None on finished videos |
| Failed video | Credits return if the video fails or is rejected before it is made |
Can you run Genjutsu on your own GPU?
Yes, with D3voz/openjutsu, the local open-source Genjutsu workflow: it runs MiniMax H3 inside ComfyUI on an NVIDIA GPU and uses SAM to track the person. Its author tested it on an RTX 4060 Ti with 16 GB and 64 GB of system RAM, where a 6-second 480p clip took about 6–13 minutes. A card with more memory, such as a 5090, leaves room for higher resolutions, but 1080p is about 2.5 times slower than the 736p default. Without an NVIDIA card, use a hosted route.
| Graphics card | NVIDIA GPU; tested on an RTX 4060 Ti 16 GB |
|---|---|
| System memory | 64 GB on the test machine; the H3 int8 models stream from RAM |
| Software | ComfyUI, the listed custom node packs and model files, then the Openjutsu app |
| Render time | About 6–13 minutes for 6 s at 480p on the test card; 1080p is about 2.5× slower than 736p |
| Licences | App: MIT. The Fix faces step uses InsightFace buffalo_l (non-commercial research only) and YOLO weights under AGPL-3.0 |
No NVIDIA card, or only a few clips to make? The generator at the top of this page needs no install: you pay per video instead of setting up ComfyUI and downloading models.
Open Genjutsu pricing and credits
Prices depend on the resolution and how many seconds of your clip you use; one or two photos cost the same. The default 10-second 480p video is 108 credits, about $2.99 at the Studio monthly rate of $0.0277 per credit.
| Model | Resolution | 5 s | 10 s | 15 s | 30 s |
|---|---|---|---|---|---|
| Seedance 2.5 | 480p | 55 ($1.52) | 108 ($2.99) | 162 ($4.49) | 322 ($8.93) |
| Seedance 2.5 | 720p | 119 ($3.30) | 237 ($6.57) | 354 ($9.81) | 708 ($19.63) |
| Seedance 2.5 | 1080p | 290 ($8.04) | 579 ($16.05) | 867 ($24.04) | 1733 ($48.04) |
Dollar amounts use the Studio monthly credit price; one-time packs cost a little more per credit. Seedance 2.5 bills both the seconds of your clip and the seconds it makes, so longer and sharper videos cost more.
See credit packs and plansWhat Open Genjutsu can’t do
Open Genjutsu here works best on short, single-shot clips of one or two clearly visible people. It is not the full openjutsu toolkit, and results vary with the clip and the photo, so check these limits before you spend credits:
- It makes up to 30 seconds per video; a longer clip needs several videos.
- It swaps people only: openjutsu’s object edits, relighting, face fixes and Montage mode are not on this page.
- It swaps at most two people per video, by photo order or by your note.
- Mirrors, reflections and very small people in the background are hard for Genjutsu-style tools and may keep the original look.
- It does not run offline, in real time or on a webcam; every video is made on our servers.
- It needs the rights to the clip and the agreement of any real person shown in it.
To cast up to four real people from portraits, use Replace Person in Video instead, and for anime, cartoon or game characters see AI Character Swap for tips on drawn characters.
Open Genjutsu FAQ
How long does one Open Genjutsu video take to create?
On this page a video usually takes a few minutes, depending on its length, resolution and the queue; you can leave the page and find it later in your history. Running openjutsu locally is slower on mid-range cards: its author reports about 6–13 minutes for a 6-second 480p clip on an RTX 4060 Ti.
Can Open Genjutsu change the voice of the original video?
No. It changes how the person looks, not how they sound: the original audio of the selected seconds is laid back under the finished video. To change the voice or language, run the finished MP4 through a separate dubbing or voice tool.
Can I use Open Genjutsu on someone else’s face?
Only with their agreement. It edits a real video with AI, so use your own face, a character you have the rights to, or a person who agreed, and do not pass the result off as real footage. Uploads and notes are checked before credits are charged, and sexual or explicit edits are refused.
Is there an Open Genjutsu app for iPhone or Android?
The open-source projects run on a computer, not as phone apps. This page works in a phone browser instead: upload a clip from your camera roll, add a photo, generate, and save the MP4, with no app to install.
Try Open Genjutsu without the install
Upload a clip and a photo, check the price, then generate.
Start an Open Genjutsu swap