BitDance-ImageNet (Diffusers)

Diffusers-compatible BitDance ImageNet checkpoints for class-conditional generation at 256x256.

Available Subfolders

BitDance_B_1x (parallel_num=1)
BitDance_B_4x (parallel_num=4)
BitDance_B_16x (parallel_num=16)
BitDance_L_1x (parallel_num=1)
BitDance_H_1x (parallel_num=1)

All variants include a custom BitDanceImageNetPipeline and support ImageNet class IDs (0-999).

Requirements

flash-attn is required for model execution and sampling.
Install it in your environment before loading the pipeline.

Quickstart (native diffusers)

import torch
from diffusers import DiffusionPipeline

repo_id = "BiliSakura/BitDance-ImageNet-diffusers"
subfolder = "BitDance_B_1x"  # or BitDance_B_4x, BitDance_B_16x, BitDance_L_1x, BitDance_H_1x

pipe = DiffusionPipeline.from_pretrained(
    repo_id,
    subfolder=subfolder,
    trust_remote_code=True,
    torch_dtype=torch.float16,
).to("cuda")

# ImageNet class 207 = golden retriever
out = pipe(
    class_labels=207,
    num_images_per_label=1,
    sample_steps=100,
    cfg_scale=4.6,
)
out.images[0].save("bitdance_imagenet.png")

Local Path Note

When loading from a local clone, do not point from_pretrained to the repo root unless you also provide subfolder=.... Each variant folder contains its own model_index.json, so the most reliable local usage is to load the variant directory directly:

from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "/path/to/BitDance-ImageNet-diffusers/BitDance_B_1x",
    trust_remote_code=True,
)

Model Metadata

Pipeline class: BitDanceImageNetPipeline
Diffusers version in configs: 0.36.0
Resolution: 256x256
Number of classes: 1000
Autoencoder class: BitDanceImageNetAutoencoder

Citation

If you use this model, please cite BitDance and Diffusers:

@article{ai2026bitdance,
  title   = {BitDance: Scaling Autoregressive Generative Models with Binary Tokens},
  author  = {Ai, Yuang and Han, Jiaming and Zhuang, Shaobin and Hu, Xuefeng and Yang, Ziyan and Yang, Zhenheng and Huang, Huaibo and Yue, Xiangyu and Chen, Hao},
  journal = {arXiv preprint arXiv:2602.14041},
  year    = {2026}
}

@inproceedings{von-platen-etal-2022-diffusers,
  title     = {Diffusers: State-of-the-art diffusion models},
  author    = {Patrick von Platen and Suraj Patil and Anton Lozhkov and Damar Jablonski and Hernan Bischof and Thomas Wolf},
  booktitle = {GitHub repository},
  year      = {2022},
  url       = {https://github.com/huggingface/diffusers}
}