Type | |
Stats | 438 |
Reviews | (33) |
Published | Dec 22, 2024 |
Base Model | |
Trigger Words | woman green dress with pink palmtrees on it |
Hash | AutoV2 BBDAB88D0D |
Sorry I tried uploading like 10 other vids, each triggered a warning lol.
Trained on 150 images from 1 concert, 512x512 for a little more than 1000 steps. Use 'green dress with pink palm trees on it' if you want that specific outfit.
For others wanting to make character loras, 30 photos is plenty. 768 is what I usually train at, havent really noticed improvement with it or with 1024x1024 but is possible with 24gb of vram. The loras on 244x244 seem not to mix well with character loras like you see more lines and seams; so I think if you're low on vram stick to photos.
https://civitai.com/articles/9967 heres an article if you want some tips.
Training on this model, I've seen varied photos do help character likeliness but I still like to make loras on specific scenes/outfits. This model also has gotten similarity of people that I've had trouble with in sdxl, pony, and flux.