| Property | Value |
|---|---|
| Base Model | Waifu Diffusion v1.4 |
| License | CreativeML OpenRAIL-M |
| Training Data | 257 artworks from Hitokomoru |
| Training Steps | 15,000 |
What is hitokomoru-diffusion-v2?
Hitokomoru Diffusion V2 is a specialized text-to-image generation model fine-tuned on the distinctive artwork style of Japanese artist ヒトこもる (Hitokomoru). Built upon the Waifu Diffusion 1.4 architecture, this model has been carefully trained using 257 high-quality artworks to capture the unique aesthetic characteristics of Hitokomoru's work.
Implementation Details
The model was trained with a learning rate of 2.0e-6 and 4 batch sizes, utilizing the Aspect Ratio Bucketing Tool for optimal image processing. It supports both standard checkpoint (.ckpt) and safetensors formats, making it compatible with popular frameworks like Automatic1111's Stable Diffusion WebUI and 🧨 Diffusers.
- Supports Danbooru-style tag prompting
- Optimized for non-square resolutions
- Compatible with multiple inference frameworks
- Advanced scheduler support (including DPMSolverMultistep)
Core Capabilities
- High-quality anime character generation
- Detailed background and environment rendering
- Support for complex compositional prompts
- Flexible resolution handling
Frequently Asked Questions
Q: What makes this model unique?
The model specializes in reproducing the distinctive art style of Hitokomoru, while maintaining the versatility of Waifu Diffusion's architecture. It excels at creating high-quality anime-style characters with detailed backgrounds.
Q: What are the recommended use cases?
This model is ideal for generating anime-style character illustrations, particularly those requiring detailed character designs and environmental elements. It works best with properly structured prompts including quality tags and detailed descriptions.