Earngenix Logo
Skip to main content

Glossary · Post-Processing

What Does Upscale Mean in ComfyUI?

By Earngenix Team ·

⚡ 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.

A latent upscale followed by a second sampling pass can add detail a plain model upscale can't — like sharper fabric texture or facial features — because the model actually generates new pixels rather than just enlarging existing ones.

Frequently Asked Questions

The underlying idea is the same — increasing resolution while reconstructing believable detail — but video upscale models have to keep that detail consistent across every frame, so they need more VRAM and specialized models rather than a plain image upscaler run frame by frame.

Yes. Both model-based and latent upscaling process a larger image than your original generation, which uses more VRAM — sometimes enough to trigger an out-of-memory error even if the initial generation ran fine.

It depends on your source: models trained on photos (like general-purpose ESRGAN variants) work best on photographic content, while models trained on illustrations or anime handle stylized art more cleanly. Check the model card for what it was trained on.

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

Discussion

Join the discussion

Sign in to leave a comment or reply

💬

No comments yet

Be the first to share your thoughts!