a artistic caricature lora with a trigger word of <code>badchar</code>. prompts in the style of <code>a badchar caricature of XXX</code> works well. I suggest starting with a weight of about 0.75. Please add images of what you manage to generate I would be most interested in seeing what everyone creates.zit v0.1: It isn't really doing the bad caricature of the original properly, but it is still interesting enough to share. It adds something interesting at a variety of strengths 0.6 will give a nice style overall, but 0.8 and above will take on the source oil painting style of the source material used to train the LoRA. I'd suggest adding "A bad char of " to the beginning of your prompt, but it's not 100% neededFlux v2.0: is much improved at creating the desired output. I recaptioned the training data and used a different training method. still, use <code>a badchar caricature of XXX</code>Flux v1.0: Didn't work as well as I had hoped. This is probably due to Flux not knowing what any famous people actually look like. It probably needs a different approach to make it work better. Until I figure out what that is here is what I have so far.

BadChar

This uses the same dataset as the SDXL version. However, I don't feel like it works as well. This is probably due to Flux not knowing what any famous people actually look like. It probably needs a different approach to make it work better. Until I figure out what that is here is what I have so far.

Files with SHA256: 4921b4b54fae29c478a4f3c9f979ed2212712d7e07ed07aec834691c1354387a