Upload folder using huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# BiliSakura/PixelFlow-diffusers
|
| 2 |
|
| 3 |
Self-contained PixelFlow checkpoints for Hugging Face diffusers. Each variant folder ships its own `pipeline.py`, component modules, and weights.
|
|
@@ -43,6 +62,10 @@ For class-conditional [`PixelFlow-256/`](PixelFlow-256/), `id2label` is embedded
|
|
| 43 |
|
| 44 |
## Demo
|
| 45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
Class-to-image:
|
| 47 |
|
| 48 |
```bash
|
|
@@ -119,53 +142,3 @@ image = pipe(
|
|
| 119 |
).images[0]
|
| 120 |
image.save("demo.png")
|
| 121 |
```
|
| 122 |
-
|
| 123 |
-
Load a **variant subfolder** (e.g. `./PixelFlow-256`), not the repo root.
|
| 124 |
-
|
| 125 |
-
## Load from the Hub
|
| 126 |
-
|
| 127 |
-
```python
|
| 128 |
-
import torch
|
| 129 |
-
from diffusers import DiffusionPipeline
|
| 130 |
-
|
| 131 |
-
pipe = DiffusionPipeline.from_pretrained(
|
| 132 |
-
"BiliSakura/PixelFlow-diffusers",
|
| 133 |
-
subfolder="PixelFlow-256",
|
| 134 |
-
custom_pipeline="pipeline.py",
|
| 135 |
-
trust_remote_code=True,
|
| 136 |
-
torch_dtype=torch.bfloat16,
|
| 137 |
-
)
|
| 138 |
-
pipe.to("cuda")
|
| 139 |
-
|
| 140 |
-
image = pipe(class_labels="golden retriever", num_inference_steps=[10, 10, 10, 10]).images[0]
|
| 141 |
-
```
|
| 142 |
-
|
| 143 |
-
Swap `subfolder="PixelFlow-T2I"` and call with `prompt=...` for text-to-image.
|
| 144 |
-
|
| 145 |
-
## Conversion
|
| 146 |
-
|
| 147 |
-
```bash
|
| 148 |
-
python scripts/convert_pixelflow_to_diffusers.py \
|
| 149 |
-
--checkpoint models/raw/PixelFlow/c2i/model.pt \
|
| 150 |
-
--config models/raw/PixelFlow/c2i/config.yaml \
|
| 151 |
-
--output models/BiliSakura/PixelFlow-diffusers/PixelFlow-256
|
| 152 |
-
|
| 153 |
-
python scripts/convert_pixelflow_to_diffusers.py \
|
| 154 |
-
--checkpoint models/raw/PixelFlow/t2i/model.pt \
|
| 155 |
-
--config models/raw/PixelFlow/t2i/config.yaml \
|
| 156 |
-
--output models/BiliSakura/PixelFlow-diffusers/PixelFlow-T2I \
|
| 157 |
-
--skip-text-encoder
|
| 158 |
-
```
|
| 159 |
-
|
| 160 |
-
## Citation
|
| 161 |
-
|
| 162 |
-
```bibtex
|
| 163 |
-
@article{chen2025pixelflow,
|
| 164 |
-
title={PixelFlow: Pixel-Space Flow Matching for High-Resolution Image Synthesis},
|
| 165 |
-
author={Chen, Shoufa and others},
|
| 166 |
-
year={2025},
|
| 167 |
-
eprint={2504.07963},
|
| 168 |
-
archivePrefix={arXiv},
|
| 169 |
-
primaryClass={cs.CV}
|
| 170 |
-
}
|
| 171 |
-
```
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: diffusers
|
| 4 |
+
pipeline_tag: text-to-image
|
| 5 |
+
tags:
|
| 6 |
+
- diffusers
|
| 7 |
+
- image-generation
|
| 8 |
+
- class-conditional
|
| 9 |
+
- imagenet
|
| 10 |
+
- pixelflow
|
| 11 |
+
- flow-matching
|
| 12 |
+
widget:
|
| 13 |
+
- text: golden retriever
|
| 14 |
+
output:
|
| 15 |
+
url: PixelFlow-256/demo.png
|
| 16 |
+
language:
|
| 17 |
+
- en
|
| 18 |
+
---
|
| 19 |
+
|
| 20 |
# BiliSakura/PixelFlow-diffusers
|
| 21 |
|
| 22 |
Self-contained PixelFlow checkpoints for Hugging Face diffusers. Each variant folder ships its own `pipeline.py`, component modules, and weights.
|
|
|
|
| 62 |
|
| 63 |
## Demo
|
| 64 |
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
Class 207 — golden retriever, 256×256, 40 steps (`[10, 10, 10, 10]`).
|
| 68 |
+
|
| 69 |
Class-to-image:
|
| 70 |
|
| 71 |
```bash
|
|
|
|
| 142 |
).images[0]
|
| 143 |
image.save("demo.png")
|
| 144 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|