>>44879093Better example.
All images: Steps: 30, Sampler: DPM++ 2M Karras, Seed: 3178764142, Size: 512x768, Model hash: dcffbff16c, Model: Gekijōban Shinseiki Dai Rantō HLL Mix 1.0 + 3.1 You Can (Not) Proompt II HD Remix &IRYS, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 15, Hires upscaler: 4x-AnimeSharp
Top left: CFG scale: 30, Mimic scale: 8, Threshold percentile: 100, Mimic mode: Half Cosine Up, Mimic scale minimum: 3, CFG mode: Half Cosine Up, CFG scale minimum: 3
Top center: CFG scale: 8
Top right: CFG scale: 30
Bottom left: CFG scale: 20
Bottom center: CFG scale: 30, Mimic scale: 5, Threshold percentile: 97.5, Mimic mode: Cosine Up, Mimic scale minimum: 3, CFG mode: Power Up, CFG scale minimum: 3, Power scheduler value: 4
Bottom right: CFG scale: 11, Mimic scale: 30, Threshold percentile: 90, Mimic mode: Cosine Down, Mimic scale minimum: 3, CFG mode: Power Down, CFG scale minimum: 3, Power scheduler value: 4
Again, do not try to do the "CFG scale lower than Mimic scale" thing on DPM SDE or 2s a unless you want to see nightmare fuel, kek. Have also noticed that turning the steps down with those two samplers helps here, since the longer they went on, the WORSE it got towards the end. Whereas with Euler or DPM 2M, it didn't start going haywire.
