vanzl/DiffusionART-SD35M-MultiReward

DiffusionART: SD3.5-M Multi-Reward

Powered by Stability AI.

A PEFT LoRA adapter for Stable Diffusion 3.5 Medium, post-trained with DiffusionART using GenEval, CLIPScore, PickScore and HPSv2 feedback.

Paper | Code | Project Page | Demo

Only the LoRA adapter is included; download the SD3.5 Medium base model separately. The adapter uses rank 32 and alpha 64.

Samples

Same prompt and seed within each pair.

Pre-trained	DiffusionART
Macro photograph of a monarch butterfly resting on purple lavender, crisp wing detail and soft garden background.
A small modern cabin beside a snowy pine forest at blue hour, warm light visible through two windows.
A red ceramic teapot to the left of a blue glass cup on a white kitchen table, natural window light.

Inference

Obtain access to the gated base model, authenticate with Hugging Face, and install the verified dependencies in an isolated Python environment:

Verified dependencies

pip install torch==2.7.0 diffusers==0.35.1 transformers==4.57.1 \
  peft==0.17.1 accelerate==1.8.1 huggingface_hub==0.36.2 \
  safetensors==0.5.2 sentencepiece==0.2.1 protobuf==5.29.5 Pillow

Use an official CUDA-enabled PyTorch wheel compatible with your driver. Then authenticate with an account that has access to the base model and this adapter.

Download the adapter and run inference:

python -c 'from huggingface_hub import snapshot_download; snapshot_download("vanzl/DiffusionART-SD35M-MultiReward", local_dir="diffusionart-mr4")'
cd diffusionart-mr4

python infer.py --prompt "A horse running through a sunlit meadow." --output image.png

The default settings are 512 x 512, 40 sampling steps, guidance scale 1.0 and seed 42. --cpu-offload is optional. The script attaches the adapter to the base transformer with PeftModel.from_pretrained.

See paired examples and inference details.

License

The adapter is a derivative of Stable Diffusion 3.5 Medium and is subject to the included Stability AI Community License and upstream acceptable-use policy. See LICENSE.md and NOTICE. The training framework's code license does not replace the model license.

Citation

@misc{wan2026diffusionartefficientstablescalable,
  title={DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models},
  author={Zhenglin Wan and Jiayu Chen and Zhengfeiyang Zhang and Pengfei Zhou and Wangbo Zhao and Fangyuan Kong and Pengfei Wan and Kun Gai and Yang You},
  year={2026},
  eprint={2610.12572},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2610.12572},
}