ourIllustrious v1 – an all-around model that excels at both realism and anime WITHOUT ANY LORAs.
Prompt quality tags:
At the start of a positive prompt, I highly recommend adding the word: newest
At the start of a negative prompt, I highly recommend adding the word: oldest
If not using Textual inversion (embeddings) in positive prompt, always append at end of positive prompt: masterpiece, best quality
If not using Textual inversion (embeddings) in negative prompt, always append at start of negative prompt: 3d, worst quality, bad quality, low quality, lowres, blurry, censor, bad anatomy, deformed
Style modifiers:
For realistic images append at end of prompt: artstation, realistic
For 2D / anime images append at end of prompt: anime screencap, anime coloring
Textual inversion (embeddings): Use lazypos (positive) and lazyneg (negative) from janxd to steer style via textual inversion embeddings. (Recomendation)
Sampler & settings:
Sampler: DPM++ 2M Karras
Steps: 30.
Guidance (CFG scale): 2.
Resolution: 1024 × 1536 or any resolution with 1024 width or height.
VAE: Skip using VAE; built-in backend already handles it.
Model base: WAI-NSFW-illustrious-SDXL v14