SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI imagesPuLID

PuLID

PuLID is an open-source face identity customization project from ByteDance, and its paper was accepted by NeurIPS 2024. It achieves pure and fast identity customization through Contrastive Alignment, and can maintain a person's ID consistency in image generation. The project provides two types of models, PuLID (based on SDXL) and PuLID-FLUX. It has officially released the v1 code and models, v1.1, and PuLID-FLUX-v0.9.1 and other versions, and supports local Gradio demos and HuggingFace online demos. Among them, PuLID-FLUX has been optimized to run on 16GB GPUs, and the local Gradio demo supports 12GB GPUs. Dependencies are Python >=3.9, PyTorch >=2.0 (when not using flux-dev-fp8) or PyTorch >=2.4.1.

Open website

About this tool

PuLID is an image generation identity customization solution proposed by ByteDance Inc. The paper is titled "PuLID: Pure and Lightning ID Customization via Contrastive Alignment" and has been accepted by NeurIPS 2024. The official code is hosted on GitHub. The authors are Zinan Guo, Yanze Wu, Zhuowei Chen, Lang Chen, Peng Zhang, and Qian He.

Main purpose: to achieve person identity (ID) customization during AI image generation, that is, to maintain the facial identity consistency of a specific person when generating images, while providing "Pure" and "Lightning" customization effects. This project targets the field of AI-driven image generation.

Main capabilities and versions:

  • Provides v1 code and models, released on 2024.05.01.
  • Released the PuLID-FLUX-v0.9.0 model on 2024.09.12.
  • Released the latest models PuLID-v1.1 and PuLID-FLUX-v0.9.1 on 2024.10.31, and updated the revised arXiv paper, including more results, details, and analysis.
  • Shared a preview of the v1.1 model on 2024.05.23.
  • Also officially recommends DreamO, a unified image customization framework that supports ID, IP, Try-On, and style tasks.
  • Supports consumer-grade GPUs: PuLID-FLUX can run on 16GB GPUs; the local Gradio demo now supports 12GB GPUs.

Usage:

  • Dependencies: Python >= 3.9 (Anaconda or Miniconda is recommended); if flux-dev-fp8 is not needed, then PyTorch >= 2.0, otherwise PyTorch >= 2.4.1 is required.
  • Provides a local Gradio demo that can be run locally.
  • Provides an online HuggingFace demo: for the SDXL version, see https://huggingface.co/spaces/yanze/PuLID；FLUX version, see https://huggingface.co/spaces/yanze/PuLID-FLUX。
  • Documentation and demos for PuLID-FLUX can be found separately.

Third-party implementations and integrations:

  • Colab:https://github.com/camenduru/PuLID-jupyter（由 provided by camenduru).
  • Replicate (PuLID):https://replicate.com/zsxkib/pulid（由 provided by zsxkib).
  • Replicate (PuLID-FLUX):https://replicate.com/zsxkib/flux-pulid（由 provided by zsxkib).
  • ComfyUI native implementation:https://github.com/cubiq/PuLID_ComfyUI（由 provided by cubiq).
  • ComfyUI implementation based on diffusers:https://github.com/ZHO-ZHO-ZHO/ComfyUI-PuLID-ZHO（由 provided by ZHO).
  • SD.Next implementation (provided by vladmandic).
  • sd-webui-controlnet integration: Mikubill/sd-webui-controlnet#2838 (provided by huchenlei).
  • PuLID-FLUX multi-person generation (combined with Regional-Prompting-FLUX, provided by Anthony).

Applicable scenarios: person identity preservation and customization in AI image generation, which can be integrated into workflows such as Stable Diffusion XL, FLUX, ComfyUI, SD.Next, and ControlNet.

Limitations and responsibility statement: The project hopes to have a positive impact on the field of AI image generation. Users are free to use this tool to create images, but must comply with local laws and use it responsibly; the developers are not responsible for possible misuse by users.

Other: If you need to report problems, you can submit an issue, or contact Yanze Wu and Zinan Guo.

Related tools

Browse category