Views
No views yet
💪 FID 1.94 🚀 Fast Speed ♻️ Low Memory Usage 🎲 Radnom Order 💡 Zero-shot Inference1from diffusers import DiffusionPipeline
2
3pipeline = DiffusionPipeline.from_pretrained("hp-l33/ARPG", custom_pipeline="hp-l33/ARPG")
4
5class_labels = [207, 360, 388, 113, 355, 980, 323, 979]
6
7generated_image = pipeline(
8 model_type="ARPG-XL", # choose from 'ARPG-L', 'ARPG-XL', or 'ARPG-XXL'
9 seed=0, # set a seed for reproducibility
10 num_steps=64, # number of autoregressive steps
11 class_labels=class_labels, # provide valid ImageNet class labels
12 cfg_scale=4, # classifier-free guidance scale
13 output_dir="./images", # directory to save generated images
14 cfg_schedule="constant", # choose between 'constant' (suggested) and 'linear'
15 sample_schedule="arccos", # choose between 'arccos' (suggested) and 'cosine'
16)
17
18generated_image.show()1@article{li2025autoregressive,
2 title={Autoregressive Image Generation with Randomized Parallel Decoding},
3 author={Haopeng Li and Jinyue Yang and Guoqi Li and Huan Wang},
4 journal={arXiv preprint arXiv:2503.10568},
5 year={2025}
6}