Stable Diffusion 3 Tiny Random
| Property | Value |
|---|---|
| Model Type | Text-to-Image Generation |
| Framework | Diffusers |
| Downloads | 45,498 |
| Base Model | Stable Diffusion 3 |
What is stable-diffusion-3-tiny-random?
This is a specialized debugging variant of the Stable Diffusion 3 model, specifically adapted from the medium version but with significantly reduced parameters and randomly initialized weights. It's designed primarily for testing and development purposes rather than production use.
Implementation Details
The model features a minimized architecture with reduced hidden sizes and attention mechanisms. Key specifications include: hidden size of 8, 2 attention heads, 2 hidden layers, and simplified VAE with 16 latent channels. The implementation uses three text encoders and a transformer, all with scaled-down configurations for debugging purposes.
- Reduced model complexity with 8-dimensional hidden states
- Minimized attention mechanism (2 heads)
- Simplified VAE architecture with 4 blocks
- Float16 precision support
Core Capabilities
- Basic text-to-image generation for debugging
- Rapid inference with minimal steps (recommended: 2 steps)
- CUDA support for GPU acceleration
- Configurable guidance scale (default: 7.0)
Frequently Asked Questions
Q: What makes this model unique?
This model's primary distinction is its intentionally minimized architecture and random initialization, making it perfect for debugging pipelines and testing infrastructure without the overhead of full-scale models.
Q: What are the recommended use cases?
The model is specifically designed for debugging, testing implementations, and development workflows. It should not be used for production or actual image generation tasks as it uses randomly initialized parameters.
