README.md

---
license: other
library_name: transformers
---


# StyleID — Stylization-Agnostic Identity Encoder
[![arXiv](https://img.shields.io/badge/arXiv-2604.21689-b31b1b.svg)](https://arxiv.org/abs/2604.21689)
[![Project Page](https://img.shields.io/badge/Project-Page-blue)](https://kwanyun.github.io/StyleID_page/)


StyleID is a CLIP-based image encoder trained to produce identity embeddings that are robust to stylization.  
It can be used for identity similarity, retrieval, evaluation, and conditioning in generative models.


<img src="https://cdn-uploads.huggingface.co/production/uploads/639d445524af4747d8d2af52/1pTEZ88YvwnbDPlV_UqpM.jpeg" width="700">

---

## Installation

```bash
pip install pillow
pip install transformers==4.52.0
```

## Usage
#### Do not use for multiple faces or faces too small to recognize.

```
import torch
from transformers import CLIPModel, CLIPProcessor
from PIL import Image

device = "cuda" if torch.cuda.is_available() else "cpu"

model = CLIPModel.from_pretrained("kwanY/styleid").to(device)
processor = CLIPProcessor.from_pretrained("kwanY/styleid")

img = Image.open(img_path).convert("RGB")
inputs = processor(images=img, return_tensors="pt").to(device)

with torch.no_grad():
    emb = model.get_image_features(**inputs)
    emb = emb / emb.norm(dim=-1, keepdim=True)
```

#### Open for non-commercial research. Do not use FFHQ for biometric human recognition