For best outputs and usage:

<<< >>>

Sampler: DPM++ SDE.

Note on the sampler: Smoothcuts XL Lightning still can use Euler a, and using it will decrease the generation time to about the half, but as back, you will need to be quite more descriptive than with DPM++ SDE, plus the visual quality will also descrease some.

Steps: 8, cfg 3, CLIP skip 2.

VAE: sdxl_vae.safetensors

Pos prompt: always start it with "score_9, score_8_up, score_7_up, score_6_up", then "source anime" (this one is optional, since it's an anime model). Also, NEVER use big weights with this model, it listens extremely well, and weights will only give you terrible outputs in 99% cases.

Neg prompt: can be left perfectly blank, no need of negative scores. Feel free to use the neg i used in my lightning version example runs: "disembodied penis, busty, ugly face, mutated hands, low res, blurry face, pumped body, athletic body, sticks, muscular female, bad anatomy, bad proportions, deformed, deformed anatomy, deformed fingers, cartoon, skinny, long torso, messy_hair, censored, lips, long neck"

<<< >>>

Sampler: Euler a.

Steps: 36, cfg 6, CLIP skip 2.

VAE: sdxl_vae.safetensors.

Pos prompt: always start it with "score_9, score_8_up, score_7_up, score_6_up", then "source anime" (this one is optional, since it's an anime model). Also, NEVER use big weights with this model, it listens extremely well, and weights will only give you terrible outputs in 99% cases.

Neg prompt: for the neg, always start it "score_6, score_5, score_4", like for example, the neg i used for my example runs: "score_6,score_5,score_4, disembodied penis, busty, ugly face, mutated hands, low res, blurry face, pumped body, athletic body, sticks, muscular female, bad anatomy, bad proportions, deformed, deformed anatomy, deformed fingers, cartoon, skinny, long torso, messy_hair, censored, lips, long neck"

What changes from v2.0 to v3.0 Trinity: input_image feature, overall quality and realism increased.

Checkpoint Merge.