Hitokomoru Diffusion
| Property | Value |
|---|---|
| License | CreativeML OpenRAIL-M |
| Training Steps | 20,000 |
| Dataset Size | 255 images |
| Model Type | Stable Diffusion |
What is hitokomoru-diffusion?
Hitokomoru Diffusion is a specialized latent diffusion model fine-tuned on the artwork of Japanese artist ヒトこもる (Hitokomoru). The model represents a significant advancement in anime-style image generation, trained using NovelAI's Aspect Ratio Bucketing Tool to handle non-square resolutions effectively.
Implementation Details
The model was trained with a learning rate of 2.0e-6 for 20,000 steps (80 Epochs) on a carefully curated dataset of 255 images from Danbooru. It comes in four training checkpoint variations (5000, 10000, 15000, and 20000 steps) and supports Danbooru-style tagging for precise image generation control.
- Supports PyTorch implementation via Diffusers library
- Compatible with ONNX, MPS, and FLAX/JAX exports
- Optimized for anime-style character generation
Core Capabilities
- High-quality anime character generation
- Support for detailed prompting using Danbooru tags
- Variable aspect ratio handling
- Specialized in both male and female character generation
Frequently Asked Questions
Q: What makes this model unique?
The model's unique value lies in its specialized training on Hitokomoru's distinctive art style, combined with advanced aspect ratio handling capabilities that allow for more flexible image generation formats.
Q: What are the recommended use cases?
This model excels at generating anime-style character illustrations, particularly those requiring specific artistic styling similar to Hitokomoru's work. It's especially effective when using detailed prompts with Danbooru-style tags.
