This model at present works with ComfyUI. I been using the trigger word, even though I don't think I need it. I feel embeddings need to go heavy on the neg prompt. I made a neg embedding, still testing it. I used it for all the demo pics.
Description
FAQ
Comments (6)
Just curious, are you hand selecting the images you are using or are you doing some sort of web scraping? I ask because i don't know this lady but when i do an image search on google it appears that there are lots of photos of her out there from after her sports illustrated career (2004 - 2015) and her looks seem to change quite a bit.
I try for around age 30 plus or minus, if you use too many images from different years it confuses the training. I do find embeddings pretty random, I can do 3 for the exact same person and they all come out different. Look younger, look older, and so on. I think its something to do with the seed it uses to make the embedding, This model came out pretty interesting not sure it totally looks like her, but I liked it. With Lora its a lot more consistent.
Use search these websites. mainly
https://mavink.com/explore.php?q=Anne+Vyalitsyna
https://yandex.com/images/search?text=Anne%20Vyalitsyna
https://www.gettyimages.ca/search/2/image?family=creative&phrase=anne%20v
Very nice. Can't wait to try it out. What are you using to train TIs on SDXL right now?
kohya_ss, unfortunately only works with ComfyUI right now. Still getting the hang of kohya, lots of options not sure what they do. Hopefully we get some photo realistic models soon.
@summer4love263 ah ok thanks
@Zovya Goddess