Velvet - a type of fabric with a dense, even pile that gives it a distinctive soft feel.
- Wiki
.
about versions:
v1 - In between of photography and artwork feel. Think over-photoshoped stock photos. Focused on looking nice.
RAW - Finetuned for a slight push into realism. More analog feel.
RA2 - More details, more contrast, more creativity. Possibly less realism.
suggestions:
scheduler: DPM++ 2M SDE
steps: 20-40
CFG: 4-6 for
avoid too much tokens in negative prompt
avoid loras (most will impose their own style, even if style is not the subject of given lora)
Description
Same as RA2, but VAE is baked.
FAQ
Comments (2)
this model is pretty great, can you merge this model with some anime model to improve on its lora combability with anime characters. a similar model was great at bring anime models to realistic effect.
https://civitai.com/models/918430/spicypizza?modelVersionId=1027981
I don't think it would work. My experience is that loras only work well with checkpoints they were trained on. Pony models in particular tend to differ a lot. So most likely effect of such merging would be losing realism and loras still wouldn't work too well. Or maybe it's just away of my capabilities. I have this merge named Rogue Pony, which is much more open for anime input, though not as much photorealistic feel (you can push it there, but a lot of prompt work is required).