Mostly works to add more detail, without altering the result too much.

Something changes. Often, the composition will be a little different. I have used a small dataset with vibrant colours and detail.

Results from this model tend to be more detailed and colorful, especially on portraits, with a weight of roughly 0.8. Adds a slight fantasy feel. Too much appears to blur the image with unnecessary details.

Experimental! I am still figuring out what it can do. Use with care at higher weights.

I've simply used FLUX images that I like and put them in a dataset. Works better for portraits. This LoRA is trained on dev-fp8.