Skip to content

Latest commit

 

History

History
28 lines (19 loc) · 1.65 KB

pulid_v1.1.md

File metadata and controls

28 lines (19 loc) · 1.65 KB

PuLID v1.1

✅ Release PuLID v1.1 for SDXL models. In this version, we have made the following improvements:

  • stronger ID editability and prompt-following ability. As shown in the examples below, pulid v1.1 can edit style, viewpoint, expression, lighting, etc., while maintaining a high degree of ID similarity.
  • better compatibility with community models. According to our tests, PuLID-v1.1 performs well on some popular and powerful base models in the community, such as Juggernaut-XL, RealVisXL, and DreamShaper-XL-Lightning.
  • better naturalness. We have noted that without any tricks, the faces generated by PuLID-v1 can be somewhat blurry and lack detail. The v1.1 version, when combined with more powerful base models, has shown improvements in facial clarity and naturalness.
  • ID similarity has beed slightly improved.

Note: The v1.1 model does not perform as well as v1 on the SDXL-lightning-4step base model, so if you want to use SDXL-lightning-4step as the base model, we still recommend using the v1 model.

Following are some examples generated by PuLID-v1.1, you can reproduce these results from our Gradio demo. release_v1 1

How to use

Local Gradio demo

python app_v1.1.py --base BASE_MODEL
Usage:
  -base: can be RunDiffusion/Juggernaut-XL-v9 or Lykon/dreamshaper-xl-lightning
         for each supported base model, we prepare some examples in the bottom of the gradio demo, please try these examples first.

Online demo

We plan to upgrade the PuLID HF demo to v1.1 soon, please stay tuned.