/vt/ - Virtual Youtubers

>>67975252
>>67975988
Yep, the only way I could achieve something good in terms of posing was by using img2img with 0.85 denoise and proper masking (not rectangles kek).
Otherwise I just get them side by side or in random poses.
Of course, img2img brings its own issues in terms of clothing and color matching because the original characters were dressed in green and red. Controlnet would be better but as this stupid workflow requires me to use stupid pipes, i have no idea where to place the controlnet conditioning.