LoHA trained on Nidalee Cosplays

Keep in mind that this is an early stage of SDXL 1.0. I know that 1.5 is still superior in many aspects, but we'll get there eventually :)

I experimented with many things, and the workflow changes on pretty much any picture. Some are done with the SD XL Refiner, while others are made with the regular SDXL model and then upscaled with a 1.5 model epiCRealism - pure Evolution V3 img2img

ComfyUI custom notes that were used sometimes:

nested notes(to make it more compact)

tomesd(for faster 1.5 model upscaling)

comfyui impact pack(wasnt really used)

SDXL Base Model, Refiner and VAE

Recommended settings:

• Make sure you are aware on the usage instructions of LORA

• VAE: sdxl_vae.safetensors · stabilityai/sdxl-vae at main (huggingface.co)

• Sampler: DPM++ SDE Karras or DPM++ 2M Karras

• Steps: 20 to 35

• Clipskip: 1

• Upscaler : 4x-UltraSharp(for 2,5D & realistic), 4xAnimesharp(for anime)

• CFG Scale : 5 to 9 (mostly 7)

• LORA weight for txt2img: 0.5-1 (for the examples i mostly used 0.8 and 1)

• Denoise strength upscaling: 0.3 to 0.55

For full body shots i recommend using: ComfyUI Impact Pack

or the a1111 alternative GitHub - Bing-su/adetailer