Remember when ComfyUI challenged the AI community to rebuild Higgsfield's popular creative tools?
Well, the joke is turning into actual software.
Meet Openjutsu, a free, open-source alternative to Higgsfield Genjutsu that lets you replace people, outfits, objects, and backgrounds in existing videos.
And the best part?
It runs locally on your own GPU.
This project is part of the broader #OpenSourceHiggsfield movement launched by ComfyUI with a rather cheeky message:
"Higgsfield said they'd open source their platform. Let's help."
And apparently, developers took that seriously.
So what can Openjutsu actually do?
1. Replace people in videos
Upload a video, provide a reference photo, and replace the original character with someone else.
The system attempts to preserve the original movements, gestures, camera angles, edits, and soundtrack.
And we're not talking about a simple face swap.
Openjutsu can replace an entire character, including their hairstyle, clothing, and body appearance.
2. Replace multiple characters
Advanced mode supports up to four characters, each with their own reference image.
Imagine taking a music video and replacing the entire band with completely different characters.
3. Edit objects and environments
Change clothing, hairstyles, props, or backgrounds.
Remove unwanted objects or introduce new visual elements.
4. Fix the results
Dedicated post-processing tools help improve facial details, correct colors, and reduce some of the artifacts introduced by video generation.
And here's how it works under the hood.
Openjutsu combines several existing AI models into one coordinated workflow:
MiniMax H3: Regenerates the video while following the original motion.
SAM 3.1: Tracks people and objects throughout the footage.
Qwen3-VL: Analyzes scenes and helps identify the correct characters.
ComfyUI: Orchestrates the entire generation pipeline.
FLUX.3 Klein: Provides additional image preparation and relighting capabilities.
The application itself is built with Next.js and uses ComfyUI as its generation backend.
But here's the really interesting part.
You don't necessarily need an expensive cloud GPU.
The developer tested Openjutsu on an RTX 4060 Ti with 16 GB VRAM and 64 GB system RAM.
It's not exactly fast.
A six-second video at 480p takes approximately 6–13 minutes on that configuration.
But it works.
And you don't have to upload your footage to a commercial video-generation service.
Of course, there are limitations.
Mirrors and reflections can confuse the system. Complex scenes may require manual corrections. Faces sometimes need additional processing.
And using reference photos of real people requires appropriate permission, especially when publishing synthetic footage.
But here's what I find fascinating about this entire movement.
We're discovering that many impressive AI applications are essentially sophisticated workflows built around existing foundation models.
The models themselves are enormously expensive to train.
But the applications built around them?
Those can often be recreated, customized, and improved by independent developers.
And with modern AI coding agents, building those applications is becoming considerably easier.
This is a pretty interesting shift in the economics of AI creative software.
A company builds a polished product around powerful models.
The community studies the workflow.
Someone rebuilds it using open tools.
And suddenly there's a free alternative you can run on your own machine.
The Openjutsu application is available under the MIT license, although the underlying AI models retain their own licenses and usage restrictions.
Maybe the future of creative AI won't belong exclusively to companies with the prettiest interfaces.
Maybe it'll also belong to developers who can reconstruct the entire experience over a weekend.
And judging by the #OpenSourceHiggsfield movement, they're already having quite a lot of fun doing it.
Openjutsu on GitHub:
https://github.com/D3voz/openjutsu
ComfyUI's Open Source Higgsfield challenge:
https://comfy.org/open-source-higgsfield/
Original demonstration:
https://x.com/laozhang1982/status/2116454591747539110
#Openjutsu #Higgsfield #ComfyUI #MiniMaxH3 #OpenSourceAI #AIVideoEditing #VibeCoding
YouTube / ABVX