While training for materials, the quality of the male faces unexpectedly improved significantly, so I'm uploading them. Please refer to the images for prompts.
Description
From 500,000 AI-generated illustrations, 3,898 were selected and trained for 7 epochs using LoRA with the LION optimizer at a learning rate of 3e-5. They were then synthesized with a 0.7 application rate.
FAQ
Comments (8)
This is a whole checkpoint for that one character?
The training images, totaling over 3000, contain a diverse range of characters with minimal repetition. If the resulting image generation model produces similar-looking faces, it could be because these features align with the current popular perception of attractive anime characters in Japan.
Wow, that's a local optimum.
Yeah, I'm confused . In simple terms is this an anime model or just a character? Both sample images look like the same guy.
@syaimu Oh I think you misunderstood. I'm not asking because of results I got. I wanted to save bandwidth and disk space, so I haven't downloaded it. I was just trying to understand what you're communicating by putting an image of two of the same character as the only examples, with no explanation.
This is not a model designed to learn and generate specific characters based on a single character. The fact that the generated outputs sometimes seem similar is likely due to using the same prompts, among other reasons. Nevertheless, they simply do not appear to be the same person to me.🫨
@syaimu Could you post some other example images using different prompts? SDXL loras are kinda big to just download on a whim, so indicating that it isn't limited to similar looking male characters would increase interest a lot.
Could you put the option to be able to use it in civitAi?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.