Earngenix Logo
Skip to main content

Glossary · Generation Modes

What Is Image-to-Image in ComfyUI?

By Earngenix Team ·

⚡ Quick Answer

Image-to-image (img2img) starts generation from an existing picture instead of blank noise. The sampler partially redraws that image based on your prompt, and a denoise strength setting controls exactly how much of the original survives.

A low denoise value makes small, subtle edits. A high denoise value lets the prompt take over almost completely, using the original image mostly as a rough starting layout.

Where You'll See It

A Load Image node feeds into a VAE Encode node, which converts the picture into the latent format the sampler works with. That latent goes into the KSampler with its denoise value set below 1.0 — a pure text-to-image workflow always runs at denoise 1.0 from an empty latent instead.

Quick Example

Load a photo of a living room, set denoise to 0.5, and prompt "mid-century modern furniture." The room's layout and lighting stay recognizable, but the furniture style shifts toward what you described.

Denoise above roughly 0.9 makes the source image almost irrelevant — at that point you're closer to text-to-image than to a genuine edit of your photo.

Frequently Asked Questions

Around 0.5–0.6 keeps the original composition and colors recognizable while still letting the prompt reshape details. Push toward 0.8–0.9 for a heavier reimagining, or below 0.3 for subtle touch-ups.

Not exactly, but staying close avoids stretching or cropping surprises. The Load Image node passes through the photo’s native size unless you resize it first, and very different aspect ratios can distort the result.

Yes — it’s one of the most common uses. Feed in a photo, set a mid-range denoise, and describe the target style in your prompt (e.g. "oil painting") to reimagine the photo in that style while keeping its composition.

See It In Action

Want to see real use cases?

Our guide covers style transfer, restyling photos, and picking the right denoise value.

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!