Home Wikituition Browse all terms Categories
Random term
Post-Processing & Retouching

AI Denoise

Noise reduction driven by a trained neural network that distinguishes noise from real detail using learned image structure.

Measured in %

In depth

Conventional denoising has to guess whether a pixel-level variation is signal or noise from local statistics alone. A trained model brings prior knowledge of what feathers, fabric and foliage actually look like, so it can suppress noise while reconstructing plausible detail — typically worth two to three stops of high-ISO performance, at the cost of long processing and occasional invented texture.

Examples in practice

An ISO 25600 wildlife frame processed with AI denoise looks close to an ISO 3200 file from the same camera.