Earngenix Logo
Skip to main content

Glossary · ComfyUI Basics

What Is Denoise Strength in ComfyUI?

By Earngenix Team ·

⚡ Quick Answer

Denoise strength controls how much of a starting image or latent gets erased and replaced during sampling. At 1.0, ComfyUI treats the input as pure noise and generates entirely new content. Lower values preserve more of the original structure — useful for editing an existing image instead of replacing it.

Think of it as an eraser-strength slider. High strength wipes the canvas nearly clean and redraws almost everything from scratch. Low strength barely touches it, only nudging small details while keeping the overall composition intact.

Where You'll See It

The denoise field sits inside the KSampler node, ranging from 0.0 to 1.0. It matters most in image-to-image workflows — in a basic text-to-image workflow, the starting latent is already empty noise, so this setting normally stays at 1.0 by default.

Quick Example

Text-to-image workflows keep denoise at 1.0, since there's no real image to preserve — just noise waiting to become one. Image-to-image workflows typically use 0.4–0.6to apply visible edits (a new style, an object swap) while keeping the original photo's layout and composition recognizable.

If an image-to-image edit isn't showing up at all, your denoise value is probably too low. If the result looks nothing like your original photo, it's probably too high.

Frequently Asked Questions

Keep it at 1.0. Text-to-image starts from an empty, fully-noised latent, so it needs complete denoising to produce a finished picture.

Almost nothing changes. At 0, the sampler barely touches the starting latent, so the output looks essentially identical to the input.

0.4–0.6 is a common starting range for image-to-image edits — low enough to preserve the original composition, high enough to actually apply visible changes. Adjust from there based on how much change you want.

See It In Action

Ready to edit a real image?

Our image-to-image guide walks through picking the right denoise value for real editing use cases.

Published: 2026-08-25 · Last updated: 2026-08-25

Discussion

Join the discussion

Sign in to leave a comment or reply

💬

No comments yet

Be the first to share your thoughts!