File size: 3,194 Bytes
a0c1743 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 | ---
license: apache-2.0
library_name: diffusers
pipeline_tag: unconditional-image-generation
tags:
- diffusers
- priming
- image-generation
- class-conditional
- one-step
- imagenet
widget:
- text: golden retriever
output:
url: demo.png
language:
- en
---
# Priming-diffusers
Native [diffusers](https://github.com/huggingface/diffusers) checkpoints for **Priming** ([Are We Overthinking One-Step Generation?](https://arxiv.org/abs/2610.20331)). Code: [JaeyeonKim01/priming](https://github.com/JaeyeonKim01/priming). Each variant folder is self-contained:
- `pipeline.py` — `PrimingPipeline`
- `scheduler/scheduling_priming.py` — `PrimingScheduler` (one evaluation; the transformer predicts the clean image)
- `transformer/transformer_priming.py` — `PrimingTransformer2DModel`
Load with PyPI `diffusers` and `trust_remote_code=True`. No extra Priming package is required. Sampling is one forward pass in RGB at 256×256 with `guidance_scale=1`. Weights are the released post-training EMA. The paper reports ImageNet-256 FID 1.38 for Priming-B and 1.19 for Priming-L.
## Available checkpoints
| Checkpoint | Path | Resolution | Steps | Guidance |
| --- | --- | --- | ---: | ---: |
| Priming-B | [`Priming-B/`](Priming-B/) | 256×256 | 1 | 1.0 |
| Priming-L | [`Priming-L/`](Priming-L/) | 256×256 | 1 | 1.0 |
## Demo
Class: *golden retriever* (ImageNet 207) — 256×256, 1 step, `guidance_scale=1.0`, `torch_dtype=bfloat16`, seed 1.
## ImageNet class labels
Each variant keeps an English `id2label` map in `model_index.json`.
- `pipe.id2label` — id → English label
- `pipe.labels` — synonym → id
- `pipe.get_label_ids("golden retriever")`
- `pipe(class_labels="golden retriever", ...)` — string labels resolve automatically
`height` and `width` default to 256. Other sizes divisible by the patch size of 16 interpolate the learned positions and rebuild 2-D rotary frequencies.
## Inference
```python
import torch
from diffusers import DiffusionPipeline
pipe = DiffusionPipeline.from_pretrained(
"BiliSakura/Priming-diffusers",
subfolder="Priming-B", # or "Priming-L"
custom_pipeline="pipeline.py",
trust_remote_code=True,
torch_dtype=torch.bfloat16,
).to("cuda")
print(pipe.id2label[207])
image = pipe(
class_labels="golden retriever",
num_inference_steps=1,
guidance_scale=1.0,
generator=torch.Generator(device="cuda").manual_seed(1),
).images[0]
image.save("demo.png")
```
Local checkout:
```python
from pathlib import Path
import torch
from diffusers import DiffusionPipeline
model_dir = Path("Priming-diffusers/Priming-B")
pipe = DiffusionPipeline.from_pretrained(
str(model_dir),
local_files_only=True,
custom_pipeline=str(model_dir / "pipeline.py"),
trust_remote_code=True,
torch_dtype=torch.bfloat16,
).to("cuda")
```
## Citation
```bibtex
@misc{park2026are,
title = {Are We Overthinking One-Step Generation?},
author = {Park, Woobin and Kim, Jaeyeon and Du, Yilun and Chen, Sitan and Han, Bohyung},
year = {2026},
eprint = {2610.20331},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2610.20331}
}
```
|