Animagine XL 4.0
| Property | Value |
|---|---|
| Developer | Cagliostro Research Lab |
| Base Model | Stable Diffusion XL 1.0 |
| License | CreativeML Open RAIL++-M |
| Training Dataset | 8.4M anime-style images |
| Training Hardware | 7 x H100 80GB SXM5 |
What is animagine-xl-4.0?
Animagine XL 4.0 represents the latest evolution in anime-focused image generation, built upon Stable Diffusion XL 1.0. This model has been extensively trained for approximately 2650 GPU hours on a massive dataset of 8.4M diverse anime-style images, with knowledge cutoff as of January 2025. The recently released Optimized version brings improved stability, better anatomical accuracy, reduced noise, and enhanced color saturation.
Implementation Details
The model employs a sophisticated tag-based training approach, utilizing state-of-the-art hardware and optimized hyperparameters. It operates with a UNet Learning Rate of 2.5e-6 and Text Encoder Learning Rate of 1.25e-6, implementing Constant With Warmup scheduling and Adafactor optimization.
- Specialized tag-ordering method for identity and style training
- Mixed precision training in FP16 format
- Batch size of 32 with gradient accumulation steps of 2
- Training resolution of 1024x1024
Core Capabilities
- High-quality anime-style image generation with precise control through tags
- Multiple aspect ratio support (from 5:12 to 12:5)
- Advanced temporal control through year-specific tags
- Comprehensive rating system from safe to explicit content
- Enhanced color accuracy and anatomical consistency
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its extensive training on anime-specific content combined with optimized tag-based prompting system, making it particularly effective for anime-style image generation with precise control over output quality and style.
Q: What are the recommended use cases?
The model excels at generating anime-style character illustrations, particularly single-character scenes. It's ideal for creating high-quality anime artwork with specific style requirements, supporting various aspect ratios and temporal styles.
