⚡ Quick Answer
Upscaling increases the resolution of a generated image or video after it's made. ComfyUI does this two different ways: a dedicated AI upscale model that enlarges the pixels and reconstructs believable detail, or a "latent upscale" that resizes the image inside the diffusion process and runs it through the sampler again to add genuinely new detail.
These two approaches solve different problems — one sharpens and enlarges what you already have, the other lets the model rethink fine detail at a higher resolution.
Where You'll See It
A model-based upscale uses a Load Upscale Model node feeding an Upscale Image (using model) node. A latent upscale instead uses an Upscale Latent node placed between two KSampler passes — this second style is often called a "hires fix."
Quick Example
Generate an image at 1024×1024, then run it through Upscale Image (using model) with a 4x upscale model. The output is a 4096×4096 version with sharper, reconstructed detail — but no new content that wasn't implied by the original.
Frequently Asked Questions
See It In Action
Want sharper output?
Our guide compares model upscale and latent upscale side by side.
Published: 2026-09-17 · Last updated: 2026-09-17
Join the discussion
Sign in to leave a comment or reply
No comments yet
Be the first to share your thoughts!
