Eva-Fusion-v2.22 Open-Source Image Generation Model - Free to Create Super Cool Neon Genesis Evangelion-Style Images

Eva Fusion V2.22

Developed by crumb

A stable diffusion-based text-to-image generation model focused on producing Evangelion-style images

Image Generation EnglishOpen Source License:Openrail #Evangelion-style generation #Low-resolution art creation #Specific style enhancement

Downloads 16

Release Time : 10/19/2022

Model Overview

This model is a text-to-image generation model developed on the stable diffusion framework, specifically designed to create artworks with Evangelion style. The model has undergone multiple rounds of training to progressively optimize generation results.

Model Features

Evangelion Style Specialization

Specifically optimized for Evangelion (EVA) style, capable of generating images that match the characteristics of this style

Multi-version Iteration

Progressively improved generation quality through three versions: v2, v2.1 to v2.22

Mixed Training Data

Combines automatically annotated and manually annotated training data to improve model accuracy

Model Capabilities

Text-to-image generation

Art style imitation

High-resolution image generation

Use Cases

Digital art creation

Evangelion-style illustration generation

Generate original illustrations with Evangelion style based on text prompts

Can produce unique artworks that match this anime style

Concept design

Mecha concept design

Quickly generate concept sketches for mecha designs

Provides design inspiration and preliminary visual references

🚀 EVA-Fusion 2.22 - You are (Not) HD

This project is for an art project of mine. It focuses on generating images with specific styles related to Evangelion using Stable Diffusion techniques.

✨ Features

Generate images in the style of Evangelion.
Trained with different models and parameters to achieve diverse results.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Model Training Details

Property	Details
Model Type	eva-fusion-v2, eva-fusion-v2.1, eva-fusion-v2.22 "you are (not) HD"
Init Model	For eva-fusion-v2: compvis 1 - 4; for eva-fusion-v2.1: eva-fusion-v2; for eva-fusion-v2.22: previous models + compvis1 - 4 + wd1.3 at various ratios
Learning Rate (lr)	For eva-fusion-v2: 2.057e - 06; for eva-fusion-v2.1: 1.313e - 06; for eva-fusion-v2.22: x
Batch Size (bs)	For eva-fusion-v2: 1; for eva-fusion-v2.1: 1; for eva-fusion-v2.22: x
Accumulation Steps (acc)	For eva-fusion-v2: 1; for eva-fusion-v2.1: 1; for eva-fusion-v2.22: x
Total Steps	For eva-fusion-v2: 8,000; for eva-fusion-v2.1: 10,000; for eva-fusion-v2.22: x
Training Samples	For eva-fusion-v2: 1,143 (captioned just "picture in the style of evangelion"); for eva-fusion-v2.1: + 560 (hand captioned); for eva-fusion-v2.22: x
Training Resolution	For eva-fusion-v2: 512x384 (4:3); for eva-fusion-v2.1: 512x384 (4:3); for eva-fusion-v2.22: x
Models Trained	For eva-fusion-v2: unet; for eva-fusion-v2.1: unet; for eva-fusion-v2.22: x

Generation Suggestions

For generating images, it is suggested to use Whatchamacallit.
In the negative prompt, it is recommended to include "blurry jpeg, multiple views, simple background, black and white reference sheet".
It is also recommended to include "in the style of Evangelion" in the prompt.

Model Limitation

The model wasn't trained on black and white character sheets, and it was barely trained on any grayscale images. However, it sometimes leans heavily towards them, and the reason is unknown.

Model File Link

The link to the ckpt file is here.

🔧 Technical Details

No specific technical details (more than 50 words) are provided in the original document, so this section is skipped.

📄 License

The project is licensed under the bigscience-bloom-rail-1.0 license.

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