|
--- |
|
license: other |
|
license_name: sdxl-license |
|
license_link: >- |
|
https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/LICENSE.md |
|
--- |
|
|
|
The technical details of Illustrious series is being updated in https://www.illustrious-xl.ai/blog ! |
|
|
|
# Illustrious XL v1.0 – High-Resolution Focused Illustration Generative Model |
|
|
|
## Overview |
|
|
|
**Illustrious XL v1.0** is a cutting-edge generative model developed by **OnomaAI**, built upon the Stable Diffusion XL architecture. Trained from the previous checkpoint, Illustrious XL v0.1, this model is designed specifically to generate stunning, high-resolution images. It achieves an unprecedented native resolution of **1536×1536** within the SDXL framework, setting new standards for detail and clarity. The model integrates natural language understanding with Danbooru-style tag-based prompting, enabling flexible and robust prompt handling. Thus, Illustrious XL v1.0 serves as an ideal foundation for creators requiring versatility without sacrificing image fidelity. |
|
|
|
## Knowledge Cutoff |
|
|
|
Illustrious XL v1.0 was trained in **July 2024**, with knowledge available up to **June 2024**. Thanks to native support of v0.1 LoRA, knowledge limits are largely mitigated. |
|
|
|
## Key Features |
|
|
|
### 1536×1536 Native Resolution |
|
- First-of-its-kind SDXL model supporting native **1536px resolution**. |
|
- Produces images with exceptional detail, sharpness, and clarity at high resolutions previously unattainable in SDXL. |
|
- Capable of handling resolutions ranging from **512x512** up to **1536x1536**, including non-standard resolutions like **1248x1824** without additional modifications. |
|
|
|
### NLP + Tag-Based Prompting |
|
- Combines advanced natural language processing with concise Danbooru tag-based prompts. |
|
- Users can employ plain English descriptions, precise tags, or both, providing greater control and nuanced image generation. |
|
|
|
### Extensive Compatibility |
|
- Compatible with various extensions and add-ons. |
|
- Supports seamless integration of LoRA, ControlNet modules, and other adaptation methods previously trained on Illustrious v0.1. |
|
- Allows users to easily mix and match enhancements for style, pose, or composition. |
|
|
|
### Pretrained Base Model |
|
- Provided as a pretrained base checkpoint without fine-tuning on specific aesthetics or biases. |
|
- Ideal for creators seeking a flexible foundation to apply their fine-tuning, such as LoRA training, for specialized art styles or outputs. |
|
|
|
## Future Plans |
|
|
|
### Higher Resolutions in v3 |
|
- Upcoming releases (Illustrious XL v3) aim to support higher resolutions, targeting up to **2k resolution and beyond**. |
|
- Anticipate further enhancements in detail and scaling, enabling the generation of ultra-crisp, large-format images. |
|
|
|
### v-Parameterization Models & Color Control |
|
- A specialized v-parameterization variant is in preparation, designed to push beyond SD XL's known limits academically and practically. |
|
- Future updates will emphasize enhanced color control, offering precise adjustments for color balance and consistency. |
|
|
|
### Continuous Improvement |
|
- The Illustrious roadmap commits to continuous integration of community feedback and incorporation of the latest research. |
|
- Efforts in high-resolution research will persist, pushing boundaries with careful and innovative methods. |
|
|
|
|
