Qwen Image Edit Lightning Lora blocks certain model capabilities, particularly related to style
Hey all. First off, thank you very much for these loras -- they make getting fast results so much easier, and are very reliable.
One thing I've noticed, between QE and QE 2509 is: the 8step loras (and I think the 4step), on 2509 seem to block QE style changes. It's fantastic for things like clothes swapping and character posing. But prompts related to holistic style changes -- 'Render this in a 3D style' -- don't seem to work with the lora.
What's really standing out is how, in the render preview on ComfyUI, the very first step seems to start off with the style change, and then snaps back to erasing the style thereafter. It was an interesting change in results, and sure enough if I disable the lora, the ability seems to come back. It's still possible with the lora on, but is much more hit and miss and requires some firmer prompting to have a chance at it.
Just reporting some results, since this is all pretty fascinating to me.
Am i wrong or is it also the model itself ?
On images like illustrations the old version can do "Change the style into a photo".
But the new one not.
the new model indeed seems to be less capable of changing styles