Hi, with this model, I want to be as near as possible to a digital painting production.

I've created it by merging my trained lora with a chekpoint merge (mainly on Babes 2.0).

Recommended settings :

• Sampler : DPM++ 2M Karras

• Steps : > 25

• CFG : 7 (don't change a lot of things)

• Hires. fix : 4X_KCJPUNK

• Denoising strength: 0.2-0.4

• Adetailer : face_yolov8n

• Clip skip : 2 seems slightly better

• No VAE needed

If you get stuck with a same position try using negative embedding, otherwise do not use them.

You can find three different version, ckp/safetensors/pruned, I prefer the ckpt one.

I'm really new to this if you find any major error or you want to provide me with some tips, be sure to send me a message !