Tried SDXL and it’s slow af. It also needs more RAM; everytime I change models my PC will freeze for a few seconds as it sucks all my RAM (I only have 16 GB, idling at 7 - 8 GB). I tried generating non-anime stuff and it seems it generated better results with less prompts compared to SD 1.5. As for anime stuff, I only tried one model and I still needed as many tags as I used on SD 1.5. I’m not sure if it looks better than SD 1.5. I didn’t try much as each generation took 4:20 minutes on average, 7x slower than SD 1.5. Btw I didn’t edit the image above at all, it’s pure txt2img.

Prompts

Made in ComfyUI

Pos: (masterpiece, best quality:1.3), 1girl standing on a cliff above sea, white sundress, from below, extremely detailed, atmospheric

Neg: embedding:negativeXL_D, (worst quality, low quality:1.3), bad anatomy, bad proportions, disfigured, amputee, disembodied limb, severed limb, missing limb, missing arms, extra arms, extra legs, extra hands, missing finger, fewer digits, extra digits, from above

Checkpoint: CounterfeitXL

Size: 1024 x 1024

Seed: 502997881848

Steps: 20

CFG: 8

Sampler: DPM++ 2M Karras