Personalize Anything

[zengels]

Hello, I just read these pages:
https://fenghora.github.io/Personalize-Anything-Page/
and
https://github.com/fenghora/personalize-anything.

They seem promising. Would it be interesting for Krita AI Diffusion? The page introduces a training‐free method for personalized image generation using diffusion transformers. Their approach, based on adaptive token replacement and patch perturbation strategies, achieves strong identity preservation without additional fine-tuning.

Considering Krita AI Diffusion—which integrates AI-driven image generation directly into Krita for tasks like inpainting, outpainting, and refining artwork—it might be very interesting to explore whether these techniques can be adapted to improve personalization and control within the Krita workflow. For instance, incorporating layout-guided generation or multi-subject personalization from this method could enhance the flexibility and fidelity of generated content in Krita. Could this be a non-redundant method to increase the potential of Krita AI?