# PixNerd-XL-16-512

Self-contained PixNerd-XL/16 checkpoint inside [`BiliSakura/PixNerd-diffusers`](https://huggingface.co/BiliSakura/PixNerd-diffusers). Runtime dependencies: this folder + PyPI `diffusers`/`torch` only.

## Hub path

`BiliSakura/PixNerd-diffusers/PixNerd-XL-16-512`

## Layout

```text
PixNerd-XL-16-512/
├── pipeline.py
├── model_index.json
├── conversion_metadata.json
├── transformer/
└── scheduler/
```

## Load

```python
import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "BiliSakura/PixNerd-diffusers/PixNerd-XL-16-512",
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
).to("cuda")

# timeshift=3.0 and order=2 are defaults in scheduler/scheduler_config.json

images = pipe(
    class_labels="golden retriever",
    height=512,
    width=512,
    num_inference_steps=25,
    guidance_scale=4.0,
).images
```

## Demo

![PixNerd-XL-16-512 demo](demo.png)

Class 207 — golden retriever, 512×512, 25 steps.