WeReCooking/Qwen-Image-Edit-CPU
3
Qwen-Image-Edit Rapid-AIO / Free CPU
Image editing and text-to-image via GGUF on free CPU hardware.
- Diffusion: Rapid-AIO v23 Q3_K (Lightning pre-fused, 4 steps) from Arunk25
- Text encoder: Qwen2.5-VL-7B-Instruct-abliterated Q3KM from mradermacher
- VAE: pigqwenimage_vae from calcuis
- Alt model: Image-2512 Q3KM (txt2img only) switchable via dropdown
- Engine: stable-diffusion.cpp (C++ inference, no PyTorch)
- Speed: ~45 min per 512x512 image on 2 vCPU
- RAM: ~14.9GB model + ~1GB activations
API
from gradio_client import Client, handle_file
client = Client("WeReCooking/Qwen-Image-Edit-CPU")
result = client.predict(
prompt="transform into anime style",
negative_prompt="worst quality, blurry",
init_image=handle_file("photo.png"),
model_choice="Rapid-AIO-v23 Q3 (edit)",
aspect_ratio="Auto (match input, max 512px)",
steps=4,
cfg_scale=2.5,
guidance=3.0,
seed=-1,
api_name="/infer",
)
print(result)MCP
{
"mcpServers": {
"qwen-image-edit": {"url": "https://werecooking-qwen-image-edit-cpu.hf.space/gradio_api/mcp/sse"}
}
}CLI
python app.py "transform into anime style" -i photo.png -o edited.png
python app.py "a cat on a windowsill" -o cat.png --model "Image-2512 (txt2img)"Credits
- Rapid-AIO by Phr00t
- GGUF by Arunk25
- stable-diffusion.cpp by leejet
- Qwen-Image by Qwen
