Analysis updated 2026-05-18
Recolor or restyle an object in a photo using a written instruction.
Move a person from one photo into a different scene while keeping their identity recognizable.
Remove an object from a photo using the slower, higher-guidance editing mode.
Combine two reference images, like a person and a scene, into one edited result.
| lbouaraba/comfyui-krea2edit | lynote-ai/humanize-text | agentic-in/elephant-agent | |
|---|---|---|---|
| Stars | 279 | 279 | 278 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | easy |
| Complexity | 3/5 | 3/5 | 2/5 |
| Audience | vibe coder | researcher | general |
Figures from each repo's GitHub metadata at analysis time.
Requires ComfyUI with Krea 2 support, the Krea 2 model itself, the Qwen3-VL text encoder, and the separate Identity Edit LoRA file.
comfyui-krea2edit is a set of custom nodes for ComfyUI, a visual tool for building AI image workflows. It turns an AI image model called Krea 2 into an image editor that follows written instructions, letting you describe a change you want, like recoloring an object or changing a scene, and apply it to an existing photo while keeping the subject's identity and likeness recognizable. The project works by feeding the source image into the model in two ways at once, once as raw visual data and once through a text-understanding component so the model can connect your written instruction to what it actually sees in the picture. This dual approach is what lets it make instruction-based edits, such as "recolor the car to matte black" or moving a person into a different scene, while keeping faces and identities consistent. To use it, you need ComfyUI already set up with support for the Krea 2 model, plus a separate Krea 2 Identity Edit add-on file that this project's nodes were built to work with. The README walks through exact wiring between nodes, and includes practical tips, like using a faster mode for most simple edits, but switching to a slower, more careful mode when you are trying to remove something from a photo rather than just change it. The project explicitly states it was trained only on safe-for-work data, was not trained on any NSFW content, and its creator does not support or endorse using it to create non-consensual, harmful, or sexual images of real people, including deepfakes. The node code itself is released under the Apache 2.0 license, while the separate AI model weights it depends on have their own license terms.
A ComfyUI add-on that lets you edit photos with written instructions while preserving the subject's identity, built for the Krea 2 image model.
Mainly Python. The stack also includes Python, ComfyUI, Krea 2.
The node code is Apache 2.0, free to use and modify including commercially, the separate AI model weights have their own license terms.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.