Update V2 v-pred
V2 is a complete new merge based on Rouwei 0.8 v-pred. For some time, I did not know how to proceed with this model and then gave mentioned Rouwei model a try and found out it worked pretty well in producing both semi-realistic Anime style and "realistic" semi-realistic pictures.
Since this model is now V-pred based, the settings changed quite a bit:
Sampler / Scheduler: Euler A + SGM Uniform or Euler Negative with Align your Steps
CFG: 3-5
Steps: 25
Resolutions: tested with SDXL resolutions
Highres fix and adetailer if needed depending on use case
VAE is included
New Quality tags:
Positive
masterpiece,best quality,amazing quality
Negative
low quality, worst quality, watermark, signature, censored, mosaic censoring, bar censor
In order to get a more realistic look, add the tags "realistic, photorealistic" to the positive prompt.
Version 1
This Semi-Realistic Checkpoint is based on Illustrious XL 1.1 and is able to produce both 2.5D Anime style and Semi-Realistic pictures.
It understands both natural language and Booru tags and supports resolutions up to 1536x1536 px.
Please see below Quality tags for each style.
Recommended settings:
Sampler: Euler A / Scheduler: SGM Uniform
CFG: 3-5
Steps: 25-30
Resolutions: works stable up to 1536x1536 px
Highres fix and adetailer if needed depending on use case
VAE is included
Quality tags for Anime Style
Positive:
masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest,
Negative:
modern, recent, old, oldest, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digits, fewer digits, cropped, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, signature, conjoined,bad ai-generated
Quality tags for semi-realistic style
Positive:
(masterpiece, best