/vt/ - Virtual Youtubers

>>39195354
>https://files.catbox.moe/vldwn1.png
Decided to give this a shot, just clicked a random link and looked through the tags/settings. A few thoughts:

- You have a mix of old emphasis style, which is [] and (()), and new style, which is (tag:1.2). If you want to convert this, it's approximately +- 0.1 for every layer of () or []. So ((student uniform)) becomes (student uniform:1.2). This is not exact and will change your image slightly but it's approximately the same amount of weighting.

- Your negative prompt is kinda wild, there's a bunch of things that could not really apply to the image ever. I've had more success in general with a smallish generic negative prompt and then specifically adding relevant things since you have more control over trying to exclude stuff. Like for example I don't know why you'd need "denim" in here because nothing in the prompt would make denim show up in the first place. It's also missing the part the anon who made the model said you should add to your negative prompts using it:
letterbox, censored, comic, (speech bubble:1.2), multiple perspectives, monochrome

- Your tag ordering is a little weird in my opinion but there's definitely a lot of rng there. I'd at least put kazama iroha before hololive.

- I've honestly never had great success with DDIM sampler for anything other than inpainting, but I'm not sure if this could really be a source of the problem. Just something to keep in mind that you can change to try and influence things

And I rolled three 3x3 grids. First is just your prompt as it was, 2nd is me changing some of the stuff I mentioned here, third is me just trying to prompt the same concept from scratch while referencing some of your tags. In the end I'm not sure it really improved that much so maybe you're just cursed and this whole post was a bit of a waste, but enjoy the bonus iroha images I guess. If I had to bet on one thing that helps it would be the hair ribbon forcing more of the top of her head into the frame.

https://litter.catbox.moe/bcuc6k.png
https://litter.catbox.moe/ja1bpc.png
https://litter.catbox.moe/7llu0b.png

Also if generation is slow for you, I'd recommend just doing everything at 512x768 at first and then come back and do highres fix on only the actually good ones since it slows down generation a lot and that helps prevent wasting time on seeds that just won't look good regardless of if you upscale them or not