No adDetailer used on showcase images so faces can be improved.
PONY version might be best at 0.9 for flexibility
The trigger "superskinnybody" doesn't make person very skinny by default, but makes it possible to create more extreme bodies.
Couldn't find a LoRA for SD1.5 that could create a very skinny body so I tried to make one myself and it went much better than expected. Trained on 100% AI generated images so total realism was not the main focus.
Breasts size can range from flat to crazy huge.
Description
Strength at 0.9 seems good, but usable from 0.7 to 1.5
FAQ
Comments (6)
Thank you for making this! Producing some weird effects on faces from my existing prompt, but I probably just need to experiment more
I didn't test it with a lora of a person, I will do that. But try low guidance, around 2
@Benkebengt Do you have any additional embeddings or tools/techniques in your generation? I tried reproducing your cover photos by dragging & dropping in comfyui and the prompt/settings look correct, but my outputs are all just grey noise. Turned off clip_skip and they aren't as noisy, but the quality isn't nearly as good as yours!
@but1er I just tried and got just noise too (I guess we talk about the pony version), The SDXL version works fine even with the pony checkpoint so this is strange. I'm using forge most of the time, and it doesn't have clip skip when using XL for some reason. In forge I get the exact the same image as original one. I don't have any embeddings but it has a lot of tools by default. I will try to figure it out.
@Benkebengt Oh, sorry I should have mentioned that! Yes, I used the PONY version. Will give SDXL a try
@but1er I have made two more images with the civitai online tool: https://civitai.com/posts/15724676
Try if you can get the workflow from one of those. One is with cfg 7 so it looks prettty bad.