Animagine Xl 3.0 Base

🚀 Animagine XL 3.0 Base

Animagine XL 3.0 Base is the foundational version of a sophisticated anime text - to - image model, focusing on establishing core functionalities for high - quality anime image generation.

🚀 Quick Start

Animagine XL 3.0 Base is the starting point of the Animagine XL 3.0 project. However, it's not recommended for direct inference. For inference, please use Animagine XL 3.0.

✨ Features

• Two - stage Development: Employs a two - stage development process based on transfer learning, which effectively addresses issues like broken anatomy in the UNet after the first training stage.
• Structured Prompting: Has a specific prompt template for optimal results, guiding users on how to structure their prompts.
• Special Tags: Trained with special tags for quality, rating, and year, providing more control over the generated images.

📦 Installation

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

💻 Usage Examples

Basic Usage

For optimal results, follow the structured prompt template:

1girl/1boy, character name, from what series, everything else in any order.

Advanced Usage

To guide the model towards generating high - aesthetic images, use negative prompts:

nsfw, 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, artist name

Prepend prompts with:

masterpiece, best quality

But be cautious when using masterpiece and best quality due to the risk of NSFW content. It's better to add nsfw, rating: sensitive to the negative prompt and rating: general to the positive prompt. Also, use a lower classifier - free guidance (CFG Scale) of around 5 - 7, sampling steps below 30, and Euler Ancestral (Euler a) as a sampler.

📚 Documentation

Model Details

Usage Guidelines

Tag Ordering

Prompting is different in this iteration. For optimal results, follow the structured prompt template:

1girl/1boy, character name, from what series, everything else in any order.

Special Tags

The model was trained with special tags for quality, rating, and year. Although it can work without these tags, it's recommended to use them for better control.

Quality Modifiers

Rating Modifiers

Year Modifier

These tags help steer the result towards modern or vintage anime art styles, ranging from newest to oldest.

Multi Aspect Resolution

This model supports generating images at the following dimensions:

Training and Hyperparameters

• Animagine XL 3.0 was trained on a 2x A100 GPU with 80GB memory for 21 days or over 500 gpu hours. The training process had three stages:
  • Feature Alignment Stage: Used 1.2m images to familiarize the model with basic anime concepts.
  • Refining UNet Stage: Employed 2.5k curated datasets to fine - tune the UNet only.

Hyperparameters

Model Comparison

Training Config

Source code and training config are available here: https://github.com/cagliostrolab/sd-scripts/tree/main/notebook

🔧 Technical Details

The training process of Animagine XL 3.0 involves a two - stage development based on transfer learning. The first stage, Feature Alignment, uses 1.2m images to introduce basic anime concepts to the model. The second stage, Refining UNet, uses 2.5k curated datasets to fine - tune the UNet. Hyperparameters such as learning rates, batch sizes, and epochs are carefully set for each stage to optimize the model's performance.

📄 License

This model is released under the Fair AI Public License 1.0 - SD.

⚠️ Important Note

While "Animagine XL 3.0" represents a significant advancement in anime text - to - image generation, it has limitations. It prioritizes concept learning over art style, is not designed for photorealistic images, may struggle with complex anatomical structures, has dataset limitations, is not optimized for natural language processing, and using high - quality tags may carry a risk of generating NSFW content.

💡 Usage Tip

Use the structured prompt template and special tags provided to have more control over the generated images. Also, be cautious when using high - quality tags to avoid NSFW content.