/vt/ - Virtual Youtubers

>>63979743

>>63979835
>>63979879
>>63980240
>>63980390
Rock jumpscare the best jumpscare, dang it
>>63981086
Adding to what both anons said, what I like to do as well is getting the base composition/gen in txt2img and using that with controlnet in the img2img step:
https://litter.catbox.moe/5sezyj.png
Sometimes you'll have to play around with the denoising. Pic below had 85% IIRC
I left the prompt empty and this came out:
https://litter.catbox.moe/ivqi1t.png
It's kinda fried but you can work with it, and if you use prompts (please do) it will definitely better. Pic related is a repost but I did it basically like this.
>>63962235
Gomen, I didn't do a proper image but this post should do the trick.

>>60741117
I controlboxed a random image I genned at 512x512 at first! That pose was more RNG than anything https://litter.catbox.moe/us33hy.png
>>60741205
>>60741417
Thanks! I'm using that trick I learned from someone here: feed img2img a flat color image and gen with a high denoise like 0.85~0.9. For this picture, though, I genned the overall composition in txt2img and then fed the output to lineart controlnet to keep the same composition. I noticed that I can go for lower denoises and keep most of the colors intact by adding that add detail lora at lower strengths (0.2?). Pic related has almost the same settings but add detail took out most of the blueish shade on Biboo's face. Her face is more detailed as well.
I do that step of feeding controlnet before genning because If I simply gen stuff in img2img, at least one third of my gens will have something weird on it such as random bars or having the character cut in half or behind something like https://litter.catbox.moe/d8k63n.png
>inpainting
https://litter.catbox.moe/5x8soy.png
I am physically incapable of doing that. I will do my inpainting reps, I swear