Decided to split my base model into realistic and anime styles because trying to both will worsen the quality of both styles.
This model allows you to create photorealistic images easily without excessive negative prompts (check the sample images).
Recommended Parameters
Dimensions: 512x768, 512x512, 768x512, 768x768
Hires: ALWAYS use hires.fix at >0.4 denoise to improve your image after you're happy with the initial composition
Sampler: DPM++2M Karras
Sampling Steps: 20-30
CFG: 7-10
Clip: 2
Negative Prompts: (low quality, worst quality:2), bad_pictures
Description
Improvements
Improved prompt following and image quality.
All sample images are done without negative embeddings! Just (low quality, worst quality:1.4) should be enough for most use cases. You can add in bad anatomy and mutant if you're getting weird limbs. Sometimes changing the seed is sufficient to fix the image!