Somethingv2 2

Developed by NoCrypt

SomethingV2.2 is an anime-style latent diffusion model improved from SomethingV2, incorporating new technologies like mbw merging, offset noise, and VAE tuning

Image Generation EnglishOpen Source License:Openrail #Anime-style generation #High-resolution restoration #Bioluminescent effects

Downloads 1,313

Release Time : 3/8/2023

Model Overview

An improved anime latent diffusion model focused on generating high-quality anime-style images, particularly skilled at handling dark backgrounds and bioluminescent effects

Model Features

Improved Image Quality

Significantly enhances generated image quality through mbw model merging, offset noise technique, and VAE tuning

Dark Effect Optimization

Particularly adept at handling dark backgrounds and bioluminescent effects, capable of generating atmospheric images

Built-in VAE

Comes with pre-installed blessed2 VAE, requiring no additional configuration

Model Capabilities

Anime-style image generation

High-resolution image restoration

Multi-character generation

Scene generation

Use Cases

Anime Creation

Character Portrait Generation

Generate high-quality anime character portraits like Hatsune Miku

Can produce detailed, expressive character images

Scene Creation

Generate anime scenes with special atmospheres, such as bioluminescent effects against dark backgrounds

Capable of creating artistically atmospheric special scenes

license: creativeml-openrail-m thumbnail: >- https://huggingface.co/NoCrypt/SomethingV2_2/resolve/main/images/thumbnail.webp tags:

stable-diffusion
text-to-image
safetensors
diffusers inference: true language:
en widget:
- text: >- masterpiece, masterpiece, masterpiece, best quality, ultra-detailed, 1girl, hatsune miku, blue hair, upper body, looking at viewer, ?, negative space, bioluminescence, bioluminescence, bioluminescence, darkness, wind, butterfly, black background, portrait, ice example_title: example library_name: diffusers

SomethingV2.2

Welcome to SomethingV2.2 - an improved anime latent diffusion model from SomethingV2

A lot of things are being discovered lately, such as a way to merge model using mbw automatically, offset noise to get much darker result, and even VAE tuning. This model is intended to use all of those features as the improvements, here's some improvements that have been made:

Can't trust the numbers? Here's some proof

Prompts

masterpiece, best quality, ultra-detailed, 2girls, upper body, looking at viewer, ?, negative space, (bioluminescence:1.2), darkness, wind, butterfly, black background, glowing,
AND masterpiece, best quality, ultra-detailed, 2girls, hatsune miku, upper body, looking at viewer, ?, negative space, (bioluminescence:1.2), darkness, wind, butterfly, black background, glowing, (blue theme:1.2)
AND masterpiece, best quality, ultra-detailed, 2girls, hakurei reimu, (brown hair:1.1), upper body, looking at viewer, ?, negative space, (bioluminescence:1.2), darkness, wind, butterfly, black background, glowing, (red theme:1.2)
Negative prompt: EasyNegative
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3452449180, Size: 816x504, Model: somethingv2_1, Denoising strength: 0.58, Clip skip: 2, ENSD: 31337, Latent Couple: "divisions=1:1,1:2,1:2 positions=0:0,0:0,0:1 weights=0.2,0.8,0.8 end at step=13", Hires upscale: 1.9, Hires steps: 12, Hires upscaler: Latent (nearest-exact)

masterpiece, best quality, hatsune miku, white shirt, darkness, dark background
Negative prompt: EasyNegative
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 72332473, Size: 504x600, Model: somethingv2_1, Denoising strength: 0.58, Clip skip: 2, ENSD: 31337, Hires upscale: 1.85, Hires steps: 12, Hires upscaler: Latent (nearest-exact)

masterpiece, best quality, hatsune miku, 1girl, white shirt, blue necktie, bare shoulders, very detailed background, hands on own cheeks, open mouth, one eye closed, clenched teeth, smile
Negative prompt: EasyNegative, tattoo, (shoulder tattoo:1.0), (number tattoo:1.3), frills
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1829045217, Size: 456x592, Model: SomethingV2_2, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 1.65, Hires steps: 12, Hires upscaler: Latent (nearest-exact), Discard penultimate sigma: True

masterpiece, best quality, hatsune miku, 1girl, white shirt, blue necktie, bare shoulders, very detailed background, hands on own cheeks, open mouth, eyez closed, clenched teeth, smile, arms behind back,
Negative prompt: EasyNegative, tattoo, (shoulder tattoo:1.0), (number tattoo:1.3), frills
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1769428138, Size: 456x592, Model: SomethingV2_2, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 1.65, Hires steps: 12, Hires upscaler: Latent (nearest-exact), Discard penultimate sigma: True

masterpiece, best quality, hatsune miku, 1girl, white shirt, blue necktie, bare shoulders, very detailed background, cafe, angry, crossed arms, detached sleeves, light particles,
Negative prompt: EasyNegative, tattoo, (shoulder tattoo:1.0), (number tattoo:1.3), frills
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3514023396, Size: 456x592, Model: SomethingV2_2, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 1.65, Hires steps: 12, Hires upscaler: Latent (nearest-exact), Discard penultimate sigma: True

Non-miku examples

Prompts

masterpiece, best quality, 1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green background, hat, hoop earrings, jewelry, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow shirt
Negative prompt: EasyNegative
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 4018636341, Size: 440x592, Model: SomethingV2_2, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 1.65, Hires steps: 13, Hires upscaler: Latent (nearest-exact)

masterpiece, best quality, landscape
Negative prompt: EasyNegative
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1334620477, Size: 440x592, Model: SomethingV2_2, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 1.65, Hires steps: 13, Hires upscaler: Latent (nearest-exact)

Recommended settings

VAE: None (Baked in model, blessed2)
Clip Skip: 2
Sampler: DPM++ 2M Karras
CFG Scale: 7 ± 5
Recommended Positive Prompt: masterpiece, best quality, negative space, (bioluminescence:1.2), darkness, dark background
Recommended Negative Prompt: EasyNegative
For better results, using hires fix is a must.
Hires upscaler: Latent (any variant, such as nearest-exact)

Recipe

Due to SD-Silicon's Terms of use. I must specify how the model was made

Model A	Model B	Interpolation Method	Weight	Name
dpepmkmp	silicon29-dark	MBW	Reverse Cosine	dpepsili
somethingV2_1	dpepsili	MBW	Cosine	SomethingV2_2 raw
SomethingV2_2 raw	Blessed2 VAE	Bake VAE	-	SomethingV2_2

Why not call it SomethingV4?

Since this model was based on SomethingV2 and there's not THAT much of improvements in some condition. Calling it V4 is just not right at the moment 😅

I am NoCrypt

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご