liveBook · Manning

Upscaler

As explained in Local Image Generation

An upscaler is an image-processing component that increases an image's spatial resolution, optionally reconstructing or adding visual detail.

An upscaler increases an image’s scale by adding pixels that were not present in the original. Stable Diffusion images often begin at 512×512, so upscaling makes them more suitable for detailed viewing, editing, wallpapers, or printing. section 7.0 section 7.1

Use an upscaler in A1111 like this:

  • Enable Hires. Fix and choose an upscaler from its dropdown. section 7.2
  • Start with Latent for a dependable default, R-ESRGAN 4x+ for photorealistic work, or R-ESRGAN 4x+ Anime6B for anime-style images. section 7.2
  • Adjust denoising strength when img2img upscaling changes the image too dramatically; lower values preserve more of the original. section 7.3.2
  • Try a different checkpoint when you want the enlarged image to adopt another visual style. section 7.3.2

Doubling latent resolution in ComfyUI

Add Node > latent > Upscale Latent By, keep the upscale method at nearest-exact, set scale_by to 2.0, and connect the original samples output to the node input. section 8.3.1

Figure 7.3 Selecting Hires. fix and viewing the configuration options.

Inspect the Hires. Fix controls before choosing an upscaler and tuning its settings. section 7.2

Figure 8.16 The Upscale Latent By node.

Inspect the ComfyUI Upscale Latent By node used to double latent resolution. section 8.3.1

How Local Image Generation covers it

How deep it goes
Explains selection, denoising, checkpoint changes, and latent-node setup, while deferring detailed differences among upscaler models.
Examples
Uses 512×512 image limitations, A1111 Hires. Fix, R-ESRGAN choices, checkpoint switching, and a ComfyUI two-times latent upscale.
What it assumes
Assumes basic Stable Diffusion image generation and introduces resolution, upscaling, Hires. Fix, and latent-scale controls.
What you can do after
Apply it via A1111 or ComfyUI to enlarge images and control style preservation or transformation.
How it connects
Connects upscaling with Stable Diffusion, Hires. Fix, img2img, checkpoints, denoising strength, latent upscaling, and VAE decoding.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Local Image Generation. The section links point to the parts of the book it draws on.