Skip to content

Latent Upscaling

Video Lecture

Section Video Links
Latent Upscaling Latent Upscaling Latent Upscaling

Description

We can get much better results by upscaling a KSampler output latent and running it through another KSampler again.

This technique is an alternative to using ESRGAN, but where it stands out the most is bringing out the best in SD1.5 based models.

Start Workflow

Latent Upscale Start

Example Prompts

  • a person standing in a lush rainforest, holding a placard with the text "Hello World!" written on it.
  • a breathtaking alpine valley at sunrise
  • modern architectural buildings with clean lines, beautiful gardens with water features, situated on the edge of a cliff, overlooking the fjords
  • A highly fashionable woman standing in front of a white wall, wearing nice shoes, looking happy, relaxed and optimistic.

Optional SD1.5 Checkpoints

In the video I demonstrate using various SD1.5 based models from earlier lessons, plus this extra Realistic Vision checkpoint.

Download it and place it into your ComfyUI/models/checkpoints folder.

realisticVisionV60B1_v51HyperVAE.safetensors

Final Workflow

The image below was created using SD1.5 based models with latent upscaling.

Drag this image into your ComfyUI to see the workflow.

Latent Upscale Final