Views
No views yet
| Subfolder | Pipeline | Task | Resolution | Model type |
|---|---|---|---|---|
PixelGen-XL-16-256/ | PixelGenC2IPipeline | class-to-image | 256×256 | PixelGen-XL/16 |
PixelGen-XXL-16-512-t2i/ | PixelGenT2IPipeline | text-to-image | 512×512 | PixelGen-XXL/16-T2I |
1BiliSakura/PixelGen-diffusers/
2├── README.md
3├── PixelGen-XL-16-256/
4│ ├── pipeline.py
5│ ├── model_index.json
6│ ├── demo.png
7│ ├── scheduler/
8│ │ ├── scheduler_config.json
9│ │ └── scheduling_pixelgen.py
10│ └── transformer/
11│ ├── config.json
12│ └── transformer_jit.py
13└── PixelGen-XXL-16-512-t2i/
14 ├── pipeline.py
15 ├── model_index.json
16 ├── conversion_metadata.json
17 ├── scheduler/
18 │ ├── scheduler_config.json
19 │ └── scheduling_pixelgen.py
20 ├── text_encoder/
21 ├── tokenizer/
22 └── transformer/
23 ├── config.json
24 ├── diffusion_pytorch_model.safetensors
25 └── transformer_jit_t2i.pycustom_pipeline=.../pipeline.py and trust_remote_code=True. PixelGen denoises directly in pixel space (no VAE).pipe.id2label — inspect id → English label correspondencepipe.labels — reverse map (English synonym → id)pipe.get_label_ids("golden retriever")pipe(class_labels="golden retriever", ...) — string labels resolved automatically
guidance_scale=2.25, Heun solver, timeshift=2.0.PixelGen-XL-16-256)1import torch
2from diffusers import DiffusionPipeline
3
4pipe = DiffusionPipeline.from_pretrained(
5 "BiliSakura/PixelGen-diffusers/PixelGen-XL-16-256",
6 trust_remote_code=True,
7 torch_dtype=torch.bfloat16,
8).to("cuda")
9
10print(pipe.id2label[207])
11print(pipe.get_label_ids("golden retriever"))
12
13generator = torch.Generator(device="cuda").manual_seed(0)
14images = pipe(
15 class_labels="golden retriever",
16 num_inference_steps=50,
17 guidance_scale=2.25,
18 generator=generator,
19).imagesPixelGen-XXL-16-512-t2i)text_encoder/ is present; otherwise downloads from the path recorded in conversion_metadata.json.1import torch
2from diffusers import DiffusionPipeline
3
4pipe = DiffusionPipeline.from_pretrained(
5 "BiliSakura/PixelGen-diffusers/PixelGen-XXL-16-512-t2i",
6 trust_remote_code=True,
7 torch_dtype=torch.bfloat16,
8)
9
10generator = torch.Generator(device="cuda").manual_seed(42)
11images = pipe(
12 prompt="A golden retriever playing in a sunny garden",
13 num_inference_steps=50,
14 guidance_scale=4.0,
15 generator=generator,
16).imagesPixelGen-XL-16-256)1from pathlib import Path
2import torch
3from diffusers import DiffusionPipeline
4
5model_dir = Path("./PixelGen-XL-16-256").resolve()
6pipe = DiffusionPipeline.from_pretrained(
7 str(model_dir),
8 local_files_only=True,
9 custom_pipeline=str(model_dir / "pipeline.py"),
10 trust_remote_code=True,
11 torch_dtype=torch.bfloat16,
12).to("cuda")
13
14generator = torch.Generator(device="cuda").manual_seed(0)
15image = pipe(
16 class_labels="golden retriever",
17 num_inference_steps=50,
18 guidance_scale=2.25,
19 generator=generator,
20).images[0]
21image.save("demo.png")| Variant | Steps | CFG scale | Solver | Timeshift | CFG interval |
|---|---|---|---|---|---|
PixelGen-XL-16-256 | 50 | 2.25 | heun | 2.0 | [0.1, 0.9] |
PixelGen-XXL-16-512-t2i | 25 | 4.0 | adam_lm | 3.0 | [0.0, 1.0] |
height and width are fixed by each checkpoint's sample_size. Custom sizes are not supported for these exports.class_labels (integer ImageNet id or English synonym).guidance_scale > 1.0 enables classifier-free guidance over a null class token.sampling_method accepts heun or euler for C2I; T2I defaults to adam_lm.noise_scale defaults to 1.0 at 256×256 and 2.0 at 512×512 when not specified.1@article{ma2026pixelgen,
2 title={PixelGen: Improving Pixel Diffusion with Perceptual Loss},
3 author={Zehong Ma and Ruihan Xu and Shiliang Zhang},
4 year={2026},
5 eprint={2602.02493},
6 archivePrefix={arXiv},
7 primaryClass={cs.CV},
8 url={https://arxiv.org/abs/2602.02493},
9}