ByteDance Releases InfiniteYou: Flexible Photo Recrafting While Preserving User Identity

news

ByteDance introduces InfiniteYou (InfU), an innovative framework based on Diffusion Transformers that enables flexible photo recrafting while preserving user identity, addressing limitations in existing methods regarding identity similarity, text-image alignment, and generation quality

ByteDance's Intelligent Creation team has recently released InfiniteYou (InfU), a new image generation framework that can flexibly generate diverse, high-quality images based on text prompts while maintaining the user's identity.

Technical Breakthrough

Achieving flexible and high-fidelity image generation while preserving personal identity remains a challenging task. InfiniteYou is an innovative framework based on advanced Diffusion Transformers (DiTs), particularly the FLUX model, designed to address limitations in existing methods regarding identity similarity, text-image alignment, and generation quality.

InfiniteYou Examples

Core Advantages

InfiniteYou demonstrates several advantages over existing identity-preserving image generation methods:

  • High Identity Similarity: The InfuseNet component directly injects identity features via residual connections, ensuring precise preservation of user identity
  • Superior Text-Image Alignment: A multi-stage training strategy significantly improves the alignment between generated images and text descriptions
  • High Generation Quality and Aesthetics: Generated images surpass existing baseline models in terms of detail, clarity, and aesthetic appeal
  • Alleviates Face Copy-Pasting: Better generates realistic facial features based on text prompts, reducing direct face copy-pasting phenomena
  • Plug-and-Play Property: Features an ideal plug-and-play design compatible with various existing methods and tools

Practical Applications

InfiniteYou technology can be applied to various creative scenarios:

  • Personalized avatar creation
  • Diverse representation for content creators
  • Virtual try-on and image transformation
  • Personalized advertising and marketing content creation

Open Source and Availability

ByteDance has open-sourced InfiniteYou's code and models on GitHub. Developers and researchers can use its code under the Apache 2.0 license, while the models are available under the Creative Commons Attribution-NonCommercial 4.0 International Public License for academic research purposes only.

ByteDance Releases InfiniteYou: Flexible Photo Recrafting While Preserving User Identity | ComfyUI Wiki