MMDv1-18
| Property | Value |
|---|---|
| License | CreativeML OpenRAIL-M |
| Author | ShinCore |
| Model Type | Text-to-Image |
What is MMDv1-18?
MMDv1-18 is an ambitious mega-merged model that combines Stable Diffusion 1.5 with 17 other carefully selected models. Created by ShinCore, this model aims to address the fragmentation of AI image generation capabilities across various platforms while improving upon SD 1.5's base limitations.
Implementation Details
The model employs a sophisticated merging methodology that combines both standard SD BLIP/CLIP and Danbooru tags for optimal performance. It requires specific technical configurations for best results, including image dimensions as multiples of 128 and particular attention to denoising strength and CFG settings.
- Supports both CLIP and Danbooru tag systems
- Requires HiRes fix enabled with specific dimensional requirements
- Recommended CFG setting of 12.5 (adjustable)
- Optimal denoising strength around 0.69
Core Capabilities
- Improved human anatomy coherency
- Enhanced creativity and detail in fore/backgrounds
- Better responsiveness to prompt engineering
- Generalist performance across multiple genres
- Two versions available: standard and toned-down (with 25% SD 1.5 remerge)
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its comprehensive merger of 18 different models, creating a versatile generalist model that improves upon base SD 1.5's limitations while maintaining broad applicability across different use cases.
Q: What are the recommended use cases?
The model is designed for general-purpose image generation, particularly suitable for users who need consistent quality across various subjects and styles. It's especially useful for indie game developers and content creators who need versatile image generation capabilities.
