Qwen-Image-2512-FP8

FP8 quantized version of Qwen/Qwen-Image-2512 using torchao.

Quantized Components

• ✅ Transformer: FP8 weight-only
• ✅ Text Encoder: FP8 weight-only

Requirements

pip install torch diffusers transformers torchao accelerate

Usage

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "Ilus-AI/Qwen-Image-2512-FP8",
    torch_dtype=torch.bfloat16,
    use_safetensors=False,
    device_map="balanced"
)

prompt = "A cat"
negative_prompt = "low resolution, low quality, distorted, blurry"

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1664,
    height=928,
    num_inference_steps=50,
    true_cfg_scale=4.0,
    generator=torch.Generator(device="cuda").manual_seed(42)
).images[0]

image.save("output.png")

Supported Resolutions

Ratio	Size
1:1	1328x1328
16:9	1664x928
9:16	928x1664
4:3	1472x1104
3:4	1104x1472