MORE Qwen Image Edit Power-ups! (Patreon)

Video - https://youtu.be/2IiCONXEpDE

Overview

I'm still loving Qwen Image Edit 2509 and so have been working hard on trying to get the very best out of the model with a whole load of workflow changes. This newly powered-up version features:

Power LoRA loader for easy multi-LoRA loading

Alternative "2x Upscale" VAE - easily make 4096x4096 images

Prompt-based masking / alternative background removal

New Image Encoding Options

I've also been playing with the Next Scene LoRA which gives some great results... even if you're not necessarily prompting for the next scene!

https://huggingface.co/lovis93/next-scene-qwen-image-lora-2509

Workflow Information

A few extra custom nodes needed this time, with the main one being the ComfyUI-RMBG pack:

Also, the 2x Upscale VAE works a treat with Qwen too!

Qwen is getting a bunch of LoRAs now, and the Power LoRA loader lets you use them with ease:

I've also switched the to the native Torch Compile node which can still be easily bypassed if your system doesn't support it.

The options are spreading thick and fast, so I've split them out into various different groups:

Remember that "Enable Images Via Prompt" uses the standard VAE, so you'll need that loader enabled if using it - even if you're not generating output using the standard VAE. Like the note says, it is possible to pick options that don't make sense ;)

Image Latents can now be added via conditioning, but personally I've found that it is still good to add them via the Qwen Prompt node as well:

Adding via the conditioning means they don't get cropped to 1024x1024, so it can help get the top/bottom/sides of certain image ratios.

The 2x Upscale VAE is pretty awesome! Even if it is mean for WAN, it works a treat on Qwen Image Edit too! Now you can easily get 4096x4096 images :)

For image masking, you can now just prompt for whatever you want:The models used will download the first time they are used, so expect a slight delay while that happens.

Enjoy!