Upscaling Techniques
The process of increasing an image's resolution while maintaining or enhancing quality. AI upscalers can add realistic detail that wasn't in the original.
Upscaling transforms low-resolution AI art into high-resolution images suitable for printing or detailed viewing.
Types of Upscaling
Traditional Upscaling
- Interpolation (bicubic, Lanczos)
- Fast but creates blur
- No new detail added
AI Upscaling
- Neural networks trained on image pairs
- Adds plausible detail
- Can enhance textures and features
SD-based Upscaling
- Uses diffusion model with img2img
- Generates new detail
- Can modify the image significantly
Popular Upscalers
| Upscaler | Type | Best For |
|---|---|---|
| RealESRGAN | AI | General purpose |
| 4x-UltraSharp | AI | Detailed enhancement |
| ESRGAN | AI | Photorealistic |
| SwinIR | AI | Quality focused |
| SD Upscale | Diffusion | Creative upscaling |
Upscaling Workflow
Basic AI Upscale
- Generate at base resolution (512, 1024)
- Apply AI upscaler (2x or 4x)
- Result: Higher resolution image
Hires Fix (SD)
- Generate at low resolution
- Upscale with latent or AI upscaler
- Run additional denoising passes
- Adds more detail than upscaling alone
Tiled Upscaling
- Process image in overlapping tiles
- Reduces VRAM requirements
- Enables very large outputs
Tips
- Generate at highest practical resolution first
- Multiple small upscales often better than one large
- Watch for artifacts in faces and text
- Combine upscaling with face restoration