VirtualFlexIlustriousXL

This is a more agile IlustriousXL type model than the hyperrealistic model prompt and allows for greater creativity while still maintaining a high level of realism. I consider it an excellent compromise between creativity and realism for what is possible or at least for what I was able to do. It remains a prototype for now but allows for high-impact images. It accepts LoRA Pony and XL well but I always recommend not overdoing it and limiting yourself to the LoRA strictly necessary to add concepts that the model does not have. The model is also quite rich in parameters. It has a tendency to veer towards NSFW content if you insert detailed descriptions of private parts. This tendency can be partially compensated for by avoiding unnecessary descriptions and inserting NSFW in the negative prompt with a high weight (1.7-2.0) if you want to create SFW content.

recommended parameters:

Sampling method: Euler A ( DPM++ 2M)

Schedule: Automatic (Karras)

Step: 25-35

Skip: 2

CFG Scale: 6-7 max 7.5

Upscaler: Reamcri oroginal

Hires steps: 20-35

Denoising strength: 0.3-0.4

Upscale: Based on the available VRAM on A1111 : (with VRAM 12GB) -> aprox. x1.7.

On ComfyUI you can use Upscale x3

It is recommended to use the VAE: fixFP16ErrorsSDXLLowerMemoryUse_v10.safetensors if your system has low VRAM