Deviant Mix

All showcase images generated without inpainting or upscaling to show raw output. It should be possible to achieve much better results using additional tools and processes. Check the suggested resources below for models used in this merge and show support for them if you like this model.

RECOMMENDED SETTINGS

Resolutions: 832x1216 / 768x1344 / 1024x1536 (Caution: Resolutions higher than these may cause distorted body proportions depending on the image composition.)

Steps: ~30

CFG Scale: 5

Sampler: DPM++ 2M (Euler Ancestral works too, though slightly less detail and sharpness)

Scheduler: SGM Uniform (Karras works too)

Upscaler / Hires. fix (Optional): 0.3 strength, 2x upscale, 4x-Ultrasharp or SwinIR_4x

ADetailer / CLIPSeg (Optional): Use as desired

RECOMMENDED BASE PROMPTS

Feel free to experiment to your own preference. Those who want to lean more heavily into the 2D style, try without any base prompt. You can still achieve quality results and it may even be preferred.

Positive:

masterpiece, best quality, amazing quality, detailed

Negative:

bad quality, worst quality, worst detail, sketch, censor, bar censor, extra fingers, extra toes, watermark, heart-shaped pupils

For a flat anime style, add "anime coloring" to positive and "realistic, photorealistic" to negative. Recommended to use booru-style tags for prompting. The Danbooru site has an extensive tag wiki that can be used as a reference. If there's no appropriate tag for what you're looking for, sometimes creating your own in the same style will get you the results you want.