| Property | Value |
|---|---|
| License | OpenRAIL++ |
| Training Data | 10M subset of LAION (>2048x2048) |
| Paper | Latent Upscaling Diffusion Model |
| Authors | Robin Rombach, Patrick Esser |
What is stable-diffusion-x4-upscaler?
The Stable Diffusion x4 Upscaler is a specialized image enhancement model designed to increase image resolution by 4x while maintaining quality and following text prompts. Trained for 1.25M steps on high-resolution images, it combines the power of latent diffusion with controlled upscaling capabilities.
Implementation Details
The model operates on 512x512 crops and implements a text-guided latent upscaling diffusion architecture. It features a unique noise level parameter that allows fine control over the upscaling process, following a predefined diffusion schedule.
- Trained on images larger than 2048x2048 pixels
- Uses OpenCLIP-ViT/H text encoder
- Implements v-objective optimization
- Supports custom noise level inputs
Core Capabilities
- 4x resolution upscaling of images
- Text-guided enhancement control
- Noise level adjustment for fine-tuning results
- Efficient processing through latent space operations
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely combines text-guided generation with upscaling, allowing users to control the enhancement process through natural language while maintaining high fidelity to the original image. The adjustable noise level provides additional creative control.
Q: What are the recommended use cases?
The model is ideal for enhancing low-resolution images, particularly in research, artistic, and educational contexts. It's specifically designed for upscaling images while maintaining quality and allowing creative control through text prompts.