Type | |
Stats | 394 26,441 431 |
Reviews | (75) |
Published | Jan 17, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 |
Trigger Words | Asuna-BA-SD1.5-Demi |
Hash | AutoV2 50A4B2CEDC |
This is my first embedding. It was trained on 98 images of Asuna in various outfits.
Do not be alarmed in the example images if you see Asunav0.2...etc, those are the same embeddings as this, just named differently.
Focusing on a single outfit would have probably led to more consistent results. However I do like that I can get most of the outfits it was trained on by adding prompts such as "white sweater", "bunny ears", "pom pom \(cheerleading\)", swimsuit, and so on. NSFW also works quite well, it is very flexible. I did notice though that negative embeddings can make Asuna's outfits slightly less consistent, so if you are going for the most canon look of Asuna, probably don't use them.
Additionally, it is almost certainly overtrained on large breasts and cleavage (maybe I shouldn't have put large breasts in the initialization text?). But playing with prompts or if needed reducing weight to the embedding seems to address that just fine.
Most of the example images were generated with MeinaMix, check out Meina if you haven't!
Any feedback and reviews would be appreciated, as well as any requests.
If I do release a version two it will focus on fixing the breast mole issue.
Email: [email protected]
Discord: demiurgex