Denoising strength

Denoising strength controls how much an image-to-image run reshapes the source image, from 0 (untouched) to 1 (source ignored).

Denoising strength (sometimes "image strength") is the core image-to-image parameter in diffusion models like Stable Diffusion. It sets how many of the diffusion steps actually start from noise versus from your input image. At 0 the output equals the input; at 0.3-0.5 the model keeps composition and pose while restyling textures and lighting; at 0.7-0.85 it preserves only loose structure; at 1.0 the input is effectively discarded and you get a fresh text-to-image generation. It is the single most important dial for controlling how faithful an edit, restyle, or upscale stays to the original.

When to use denoising strength

• Restyling a photo or sketch while keeping its composition (0.35-0.55).
• Iterating on a generation you almost like — nudge low (0.2-0.4) to refine without losing it.
• Turning a rough layout or pose reference into a finished render.

When not to use denoising strength

• Pure text-to-image with no source — the parameter has no effect.
• When you need an exact-pixel edit; use inpainting with a mask instead.

Example

Input: img2img, source: phone photo of a living room, denoising 0.45, prompt: "cozy scandinavian interior, warm light"
Output: Same room layout and furniture placement, restyled with brighter wood tones and softer light.

Drop to 0.3 if walls/furniture start drifting; raise to 0.6 for a bolder restyle.

Common mistakes

• Setting it near 1.0 and wondering why the source image disappeared.
• Expecting low strength to fix bad prompts — structure is kept, but quality still tracks the prompt.
• Comparing values across tools blindly; UIs label and scale this differently.

FAQ

Related terms

• Seed — A seed is an integer that initializes the random number generator inside an image, video, or audio model, making generation reproducible.
• CFG scale (classifier-free guidance) — CFG scale controls how strongly a diffusion image model follows its text prompt — higher values stick closer to the prompt, lower values explore more.
• Diffusion model — A diffusion model is a generative neural network that creates images, video, or audio by iteratively denoising random noise toward a learned target distribution.
• Negative prompt — A negative prompt is text that tells an image, video, or audio generator what to avoid producing — the opposite of the main prompt.
• Outpainting — Outpainting generates new, coherent image content beyond the original canvas edges, extending a picture outward.

Sources

• Hugging Face Diffusers — image-to-image

Last updated: 2026-06-02. Raw markdown: https://promtable.com/glossary/denoising-strength.md.