Local Qwen Image Lightning for all (8 steps)

Hi,

You were waiting for a qwen extension for ForgeWebUI but it has been complicated to integrate it without breaking the rest. So CHATGPT and I did a stand alone UI using Qwen Image and Qwen Lightning lora (8 steps only) and Deepbeepmeep memory management for the GPU Poor.

So I have tested it on a 3090 with 24Gb VRAM and 64Gb RAM but it should work on 6Gb VRAM/16Gb RAM also. It will detect your hardware and set the profile automaticly.

On my setup, a 1024x1024 picture/8 steps takes between 17s to 20s like this one:

How to run it:

Download the archive, unzip it in a folder. Run Install.bat. After you can run run_app.bat when it tells you open in your browser the 127.0.0.0:7860

At the first generation, it will download the model, it's long (64Gb) but after, it's fast!

Now it's working, I will improve and adjust a couple element (4 steps, manual seeds, other lora, ...) then will work to integrate it in ForgeWebUI without breaking everything.

Enjoy!

PS : use divisible by 16 resolution, if not you'll get a message in the console and press a key, it runs anyway but it makes that pause.