Anmokomergetest1 / README.md
LieDeath's picture
Add previews and full descriptions
c189187
metadata
license: cc-by-nc-4.0
language:
  - en
library_name: diffusers
pipeline_tag: text-to-image
tags:
  - art

Anmokomergetest1

Anmokomergetest1 is a merged stable diffusion model specializes in anime, and can generate vivid and detailed also wide range anime pictures in high quality even with simple descriptions.
It is an authorized copy of @路费不够 private model in a Chinese AI community NovelAI 中文频道, and merged from 3 models: Anything-V3.0(Linaqruf), momoko-e(anonymous), mergetest1(another private model from the community).
Better performace at 1k original resolution, with Hires.fix it can produce larger pictures with more logical structures.
Provide in 2 versions, uncompressed checkpoint and compressed safetensors. The CKPT needs to work with individual VAE or it wouldn't vivid, better for merging, and the safetensors has baked in VAE, better for generating pictures.

No commercial usage!

Preview

Hatsune Miku, upper body
13261-2902195416.png
Kagamine Rin, upper body
13260-3216884448.png
Hakurei Reimu, upper body
13262-1302540835.png
Tohsaka Rin, upper body
13273-967718540.png
Misaka Mikoto, upper body
13268-1088153595.png
Yuuki Asuna, upper body
13266-2752671155.png

Usage

Use as normal stable diffusion model package v1.x, no external yaml config is needed.
Recommand settings: Steps: 12-28, Sampler: DPM++ SDE Karras, CFG scale: 5-11, Resolution: 1024x1024

Tags

Positives, less is more. Negatives, less is more, too.
Recommand Positives for Quality Control: masterpiece, best quality The other prompts as you like.
Recommand Negatives: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artists name

Bias

Low quality at too low or too high resolutions!!!
Bad eyes at low resolutions. Malformed limbs often appear. Training set Contain copyright images.

Formula

Round1 Anything-V3.0(Linaqruf)+momoko-e(anonymous) sum rate0.25
Round2 ()+mergetest1(Private, from a Chinese AI community NovelAI 中文频道) sum rate0.5
Get Anmokomergetest1.ckpt, which better for further merges.
After baked in vae-ft-mse-840000-ema-pruned(StabilityAI) VAE, pruned ema, compressed to FP16, get Anmokomergetest1.safetensors. Better for generate images.