🙏👍*** If you like it, please like it! ***👍🙏
A model of Shirogane Sama, a well-known cosplay model from Instagram.
The goal is to capture as much of the person as possible with my LORA. With little or no prompting the full body should appear. To ensure the face doesn't get lost, I make sure to train 30-40 high quality close-ups from varying angles and with varying expressions.
Flux 1D
Euler, Beta
15 - 40 steps
1.5 - 7.0 CFG
Please note I am using Euler with Beta. My go to is typically 1.5/15 to 3.5/35 CFG/steps depending on the complexity. See my images for more generation data.
Description
FAQ
Comments (4)
I reviewed your dataset, did you make the captions manually?
I used CivitAI's lora generator, which has an option to autogenerate prompts for each image. *However,* I also went through every single caption, spending more hours than I care to admit, updating and tailoring them to the scene. The generator was great at describing her features and background, but it struggled with actions and costumes. If you saw my dataset and images I used, you'll understand that meant I had to update every caption except for maybe a handful.
Looks great! Any plans for other cosplayers or maybe updates/recreations in other models like PonyRealism? Ether way great work!
Definitely planning more in this sphere, and will update when I can. Thanks!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.