yeah the artist has a really cool style and the otamatone characters are pretty, you could definitely search for tutorials on better model making, tags and such (I think the otomatone boys would definitely need tags like humanoid, object head and such, also making a clear distinction between the two like you did), hope you can do it soon cuz they are awesome

@anonimous1212comeback I'm not really sure, I haven't found an answer either, I'm confused on that. From what I know, your data/images should not be very similar though, like for example, each image you use only has a white background strictly. If I know correctly, making a model needs images that are different of the character, even including the backgrounds they are in, so that they are not hard "burned" into the model. Like for example, if you make a model with different images of the character that only have a strictly white background behind, the model may push for more white backgrounds when generating afterwards, so it's basically "burned" into the model, which is a thing to avoid.

Based off of that, I would guess that the best thing to do is just have your images having different backgrounds as much as possible (like out of your 25 images at least some have different backgrounds, also no other characters in the background if you're making a model of a specific character only). Ensuring that the images are high quality too is important too, as images with too much pixels can kind of "corrupt" the model, or just basically have worse generation results.

I suggest you search more about this through tutorial articles and by asking through the discord server, as I myself don't really know much about making the models sadly. I have heard some people learnt by only asking for guidance from discord, and someone guiding through them, so who knows. If you want, a decent article I had found is by someone called Haxivus100, they had a fairly simple explanation for character model making. It covers mostly the basics, but it's useful to get a first understanding. So yeah, I suggest you search about tips in other articles and stuff too, if you want to understand it more (since I really like these otamatone boys, and also the artist's style, if you learn how to make styles too, which disclaimer, I have heard that's also a bit different than making character models, you can search about that too). Good luck with whatever you want to do, and sorry for the yapping lmao. ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯ (Found the article, here, https://civitai.com/articles/11378/creating-a-char-for-dummiesin-illustrious).

Oh and forgot to say, it's also important to have images of the characters in different poses and views as much as possible, like in front, from behind, above, below, etc. That way, a specific pose or view isn't "burned" into the model, which again, is a thing to avoid, because then it generates the character with only specific views in mind, and for example won't be able to generate how the character would look from behind, if you wanted, but didn't use an image that shows how they actually look from behind. In general, the idea is to have images that somehow have something different, and not same for each image, whether that be poses, views, expressions/faces, etc. If you want you can ask me anything else in DMS. I don't make models, but I have read a few articles and kind of understood some basic things. Otherwise as I said, learn more from articles and from people through the server maybe, hope I somehow helped!

@anonimous1212comeback Hm, if you're talking about generating images after you publishing the model, then that means that you did something wrong while training or used images that were not very "healthy" for the training data, for example low quality pixelated images, or you didn't tag each image properly before using them for training, etc. And during generating images, after the model is published, I recommend you don't use weights over 1 (I did a little snooping and noticed that many of your images use the character and style models with weights over 1), because it's just not advised. A model should be good enough at 1 or maybe a bit less, more than 1 weight is not necessary, and personally I never had good results with over 1.

Also, you should avoid using the basic checkpoint models, like how you sometimes use Illustrious-XL. Instead use Wai-NSFW Illustrious (I prefer the 12.0 version but 13.0 is good too), or HomoSimile, which is really good for male characters. Other than that, it's also a matter of using prompts in your generation for quality (like the typical masterpiece,best quality,amazing quality / bad quality,worst quality,worst detail,sketch,censor, but there can be more too) and knowing how to construct your prompts, what words to use etc, which you can learn by observing others prompts and what works for them.

I also noticed you use lots of parethenses for lots of your prompts, or things like :1.2, etc, and although I know some people can use those and make them work, personally I don't really use them, because I have read that using them makes your other tags be a bit avoided (it's like, if you have two words for example, and one is automatically at 1, and you put the other for 1.2, the word with automatically 1 weight only may get a bit overshadowed by the other word with 1.2). Sorry if I confused you or anything, or said things that you already know, I'm just trying to help however I can and know, even if I'm yapping lol.