README.md

<p align="center">
  <img src="https://lf3-static.bytednsdoc.com/obj/eden-cn/nuhojubrps/banner.png" alt="BAGEL" width="480"/>
</p>

<p align="center">
  <a href="https://bagel-ai.org/">
    <img
      src="https://img.shields.io/badge/BAGEL-Website-0A66C2?logo=safari&logoColor=white"
      alt="BAGEL Website"
    />
  </a>
  <a href="https://arxiv.org/abs/2505.14683">
    <img
      src="https://img.shields.io/badge/BAGEL-Paper-red?logo=arxiv&logoColor=red"
      alt="BAGEL Paper on arXiv"
    />
  </a>
  <a href="https://huggingface.co/JiaxinGe/Diffusers-BAGEL">
    <img
      src="https://img.shields.io/badge/BAGEL–Diffusers-FF6C37?logo=huggingface&logoColor=white"
      alt="BAGEL Diffusers Pipeline"
    />
  </a>
  <a href="https://huggingface.co/ByteDance-Seed/BAGEL-7B-MoT">
    <img
      src="https://img.shields.io/badge/BAGEL-Model-yellow?logo=huggingface&logoColor=yellow"
      alt="BAGEL Original Model"
    />
  </a>
  <a href="https://demo.bagel-ai.org/">
    <img
      src="https://img.shields.io/badge/BAGEL-Demo-blue?logo=googleplay&logoColor=blue"
      alt="BAGEL Demo"
    />
  </a>
</p>

# BAGEL: Diffusers Integration

This repository hosts the **BAGEL** custom pipeline for 🤗 Diffusers, enabling seamless text-to-image, image editing, and visual understanding tasks with the BAGEL model.

## 🚀 Quick Start

```python
pipe = DiffusionPipeline.from_pretrained(
    "JiaxinGe/Diffusers-BAGEL",
    custom_pipeline="JiaxinGe/Diffusers-BAGEL",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)
pipe = pipe.to("cuda:0")
# (1) text→image
prompt = "A female cosplayer portraying an ethereal fairy or elf, wearing a flowing dress made of delicate fabrics in soft, mystical colors like emerald green and silver. She has pointed ears, a gentle, enchanting expression, and her outfit is adorned with sparkling jewels and intricate patterns. The background is a magical forest with glowing plants, mystical creatures, and a serene atmosphere."
out = pipe(text=prompt)
print(out)
out['images'][0].save("bagel_text2img.png")

# 2) text→image with “think”
out = pipe(
    text="a car made of small cars",
    think=True
)
print(out['text'])
out['images'][0].save("bagel_text2img_think.png")

# 3) image editing
from PIL import Image
img = Image.open("~/Bagel/test_images/women.jpg")
out = pipe(
    image=img,
    text="She boards a modern subway, quietly reading a folded newspaper…"
)
out['images'][0].save("bagel_img_edit.png")

# 4) image editing + think
img = Image.open("~/Bagel/test_images/octupusy.jpg")
out = pipe(
    image=img,
    text="Could you display the sculpture that takes after this design?",
    think=True
)
print(out['text'])
out['images'][0].save("bagel_img_edit_think.png")

# 4) image understanding
meme = Image.open("~/Bagel/test_images/meme.jpg")
out = pipe(
    image=meme,
    text="Can someone explain what’s funny about this meme?",
    understanding_output=True
)
print(out['text'])
```

## 🔧 Inference Hyperparameters

* `cfg_text_scale`: text guidance strength (typical: 4.0–8.0)
* `cfg_image_scale`: image guidance strength (1.0–2.0)
* `cfg_interval`: fraction of steps to apply CFG (e.g. 0.4–1.0)
* `num_timesteps`: total denoising steps (e.g. 50)
* `timestep_shift`: offset of denoising schedule
* `cfg_renorm_min` / `cfg_renorm_type`: renormalization settings for CFG

For detailed explanations, see the original [BAGEL](https://github.com/bytedance-seed/BAGEL) repository.