OpenJutsu locally replaces up to four characters in a single video

A cloud video editor is convenient right up until you have to upload a client's source files, an unreleased clip, or footage featuring real people. OpenJutsu offers another route: replacing characters, clothing, and objects directly on your own machine. The project reproduces one of the main Higgsfield Genjutsu workflows, but assembles it from open components and runs it through ComfyUI.

Its practical value goes beyond being a free alternative to a commercial service. OpenJutsu gives developers a ready-made architecture for a modern AI video editor: one model analyzes the scene, another tracks the selected object across frames, a third redraws the image, and a processing graph manages the entire sequence. This kind of example is more useful than yet another polished demo video: it can be dissected, modified, and integrated into a product.

How it works in a typical example

Imagine a music video featuring four performers. The user uploads the video, assigns a separate photo to each person, and starts processing. OpenJutsu must preserve the original movements, gestures, camera work, editing, and soundtrack while redrawing the appearance of the selected people. This is not limited to face replacement: the system can change a person's hairstyle, body type, and entire outfit.

The same pipeline also supports narrower edits. It can redress a character, replace a prop, remove an object from the frame, or change the background. After the main pass, separate tools repair poorly rendered faces and soften overly harsh color grading. That is an important detail: video generation rarely ends with a single button press, and OpenJutsu includes a finishing stage from the outset.

Under the hood, the work is divided among several components. According to the project description, MiniMax H3 handles video redrawing, SAM 3.1 tracks people and objects, Qwen3-VL interprets the scene's contents, and ComfyUI connects the operations into a single graph. For developers, this is a convenient entry point: each block can be studied and replaced independently without rewriting the entire interface and pipeline.

Local does not mean lightweight

OpenJutsu's main advantage is that source videos do not have to be sent to a commercial service's server. This reduces the risk of confidential material leaking and provides greater control over models, parameters, and results. According to its authors, the project runs on an RTX 4060 Ti with 16 GB of VRAM and a computer with 64 GB of RAM. Such a configuration is attainable for a small studio or developer, but it can hardly be called mainstream yet.

There is a more conspicuous limitation as well: processing is slow. The open project should be viewed as a working prototype and a foundation for experiments, not as a guaranteed replacement for the cloud product. Quality will depend on the source video, occlusions, camera movement, lighting, and available memory. The claimed capabilities should also be tested on your own footage: the project's publication and demo videos do not yet prove stability across every scene.

What developers can do with it

OpenJutsu is worth studying for teams building video editors, ad-localization tools, virtual try-on systems, or content-production software. The repository provides a clear map rather than a finished business: segmentation and tracking, scene analysis, generative redrawing, defect correction, and orchestration. Teams can adopt the entire pipeline for a prototype or test just one component on their own data.

For founders, the project is a reminder of a less pleasant reality: an interface built on other companies' models protects a product less effectively than proprietary data, speed, output quality, and a refined workflow. The community can reproduce a prominent feature relatively quickly when it is based on accessible models. The practical conclusion is straightforward: developers should test OpenJutsu as a local laboratory, while product teams should assess how much of their advantage will remain once the basic pipeline becomes available to everyone.