Note: V2 has a larger dataset than the original V1. Generally speaking, it should work better and be more stable overall. I may do a separate V3 when the fourth season airs (likely, though, I'll just do a separate S4 model of her instead), but this is probably about as good as it will get for now.
Obligatory self-promotion: consider supporting or commissioning me at https://www.patreon.com/MetalChromeX or http://www.fiverr.com/s/VY34o4m so I can afford an Nvidia 5090 when I can buy one priced at or near the MSRP (screw you, scalpers!)!
Original description:
Someone provided me with 14 fan art images (I gathered 3 of the official outfits) of Emilia's Season 4 outfit that they gathered and asked if I could try making a style model. This model works, but just barely due to the small dataset that didn't have much variety of poses, angles, etc. However, you can still at least get the outfit on other characters, if nothing else! Most of the time, the white cloak and the purple suit will generate without any issues; the keywords white cloak and purple bodysuit seem to help get it in more consistently.
When season 4 airs, I plan to incorporate it into my Emilia Illustrious model; I will also update Emilia's model soon to include a full season 3 model (edit: done!) and a Season 1-3 model with all outfits later.
Description
FAQ
Comments (3)
Most of the images I generate with LORA show me with the hood in my hand. Is it possible that another version could solve this problem?
Yeah, so unfortunately, the dataset for this is quite limited and can be very hit-or-miss. Are you using V1 or 2? V2 does work better, though you'll still need to play around a bit to get it working. I'd maybe suggest using + adding weight to keywords like 'arms to the side' or 'pointing at the viewer' or something along those lines.
@metalchromex I'm using version 2 and I'm having that problem. I'll try some prompts to see then. I'll leave that detail so you can solve it in a future version.