---
title: QwenImageEditLite
emoji: 🖼️
colorFrom: pink
colorTo: red
sdk: gradio
sdk_version: 6.14.0
python_version: "3.12"
app_file: app.py
pinned: false
---

# Qwen Image Edit Lite ZeroGPU Demo

This Space runs `Qwen-Image-Edit-2511` on Hugging Face ZeroGPU with the bundled ComfyUI runtime under `vendor/comfyui`.

## Model sources

Base models are resolved from Hugging Face URLs in [config.ini](./config.ini):

- `diffusion_models`: Comfy-Org `qwen_image_edit_2511_fp8mixed.safetensors`
- `vae`: Comfy-Org `qwen_image_vae.safetensors`
- `text_encoders`: Comfy-Org `qwen_2.5_vl_7b_fp8_scaled.safetensors`

LoRAs are also defined as Hugging Face `resolve` URLs. They are downloaded on demand into `models/loras`.

## HF token

Some LoRAs may require authentication.

- Register `HF_TOKEN` in the Space `Variables and secrets`.
- The app reads the token from the environment for Hugging Face downloads.

## Attention backend

- `attn_mode = xformers`
- No extra flash-attn or alternate attention backend setup is included.

## Notes

- `benchmark/` is intentionally omitted in this demo repo.
- Downloaded model artifacts are cached under `models/`.
- This repository includes copied ComfyUI runtime code, so keep the existing license files with the repo.