ToTheBeginning/PuLID

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3550 stars · 268 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

PuLID is the official PyTorch implementation of a NeurIPS 2024 method for inserting a specific person's identity into text-to-image generation via contrastive alignment, supporting SDXL and FLUX models. It ships pretrained models, local Gradio demos, and HuggingFace/Replicate demos, running on consumer GPUs with as little as 12-16GB VRAM.

Use cases

• generate images of a specific person from a reference photo
• preserve face identity in AI portrait generation
• customize FLUX or SDXL outputs with a consistent character
• create consistent avatars across multiple generated images
• run identity-preserving image generation on a consumer GPU
• reproduce PuLID research results

When to choose

• you need high-fidelity face identity preservation in text-to-image generation
• you want a research-backed method with pretrained checkpoints for SDXL or FLUX
• you have a modest GPU (12-16GB) and want local inference
• you want to try the method quickly via HuggingFace or Replicate demos

When to avoid

• you need multi-subject or try-on customization - consider the authors' DreamO framework instead
• you have no GPU available for local inference
• you need a production API service rather than a research codebase
• you work with non-face subject customization

Facets

library · maturity active

image-processing machine-learning stable-diffusion deep-learning artificial-intelligence image-processing computer-vision python face-customization id-preservation text-to-image flux gradio-demo research-paper gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="ToTheBeginning/PuLID")