Min-Illust-Background-Diffusion
| Property | Value |
|---|---|
| License | CreativeML OpenRAIL-M |
| Author | ProGamerGov |
| Framework | Stable Diffusion v1.5 |
| Training Iterations | 2250 |
What is Min-Illust-Background-Diffusion?
Min-Illust-Background-Diffusion is a specialized fine-tuned version of Stable Diffusion v1.5, trained specifically on artistic works by Sin Jong Hun. The model excels in generating minimalist landscape backgrounds with a distinctive artistic style, achieved through careful training with prior-preservation loss and advanced VAE implementation.
Implementation Details
The model was trained for 2250 iterations using a batch size of 4, implementing full precision training through ShivamShrirao/diffusers. It utilizes Stability AI's 1.5 MSE VAE and incorporates 4120 regularization images for maintaining consistency. The training process included train-text-encoder feature and prior-preservation loss techniques.
- Trained with full precision and prior-preservation loss
- Implements Stability AI's 1.5 MSE VAE
- Uses 4120 regularization images with "artwork style" prompt
- 50 DDIM steps and CFG of 7 for regularization
Core Capabilities
- Generates artistic landscapes in Sin Jong Hun's style
- Produces minimalist background illustrations
- Activates style using "sjh style" token in prompts
- Optimized for landscape generation rather than portraits
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in recreating Sin Jong Hun's distinctive artistic style, particularly for landscapes and backgrounds, using a specialized token "sjh style" to activate the effect.
Q: What are the recommended use cases?
The model is best suited for generating artistic landscapes and backgrounds. It's specifically noted that it may not perform well for portraits or character-focused images, as the training data primarily consisted of landscapes.
