What is Upscaling?
The process of making an image sharper than its original quality is called upscaling.
Image Upscale Workflows
Simple Upscale
The simplest and fastest upscale workflow. Choose a model based on the type and purpose of the image you want to upscale.
- RealESRGAN_x4 and 4x-Ultrasharp models are good starting points to try first.
- Upscaling too aggressively from low resolution (~1024p) may produce artifacts in fine details.
Ultimate SD Upscale
A workflow that combines the Image-to-Image process with upscaling. Through tiling, the image is divided into smaller sections and sequentially regenerated, enhancing the image by filling in lacking details.
- Adjust settings like a standard KSampler node, but keep
denoiseat 0.5 or below to preserve the original composition. - Since upscaling is done through regeneration, prompts also have an effect β make sure to write prompts that match the subject and purpose.
SeedVR2 Upscaler
A recently popular open-source upscaler that offers a good balance between upscaling and image enhancement.
- The
tile_sizemust be set smaller than the original image for upscaling to work properly.
Recommendations by Use Case
| Purpose | Recommended Model | Reason |
|---|---|---|
| General high quality | SeedVR2 Upscaler | Balanced between preserving the original and enhancing the image |
| Quick upscale | RealESRGAN_x4, etc | Fast and simple upscaling |
| Image enhancement (regeneration) | Ultimate SD Upscale | Fixes detail issues through regeneration |