Generation controls

    Denoising strength

    Also called Transformation strength, Change strength.

    Denoising strength decides how much of your input picture gets thrown away before regeneration — low keeps it nearly intact, high keeps only the general shape.

    When you hand a model an existing image, it does not start from pure noise. It adds noise to your picture and then denoises back out. Strength is how much noise it adds: at a low value the original is still clearly visible under the static, so the model has little freedom; at a high value it is almost gone and the model is effectively generating fresh with a vague memory of the layout.

    That gives the setting a very practical reading. Small values are corrections — grade shifts, texture cleanup, a slightly different expression. Middle values are restyles that keep the composition. High values are new pictures that happen to share a silhouette.

    Identity is the thing that breaks first. Faces stop being the same person well before layout stops being the same layout, so if a specific person has to survive the edit, the workable band is much narrower than it looks.

    In practice

    • Bracket it: run the same edit at low, medium and high and pick, rather than guessing once.
    • It interacts with step count — very few steps at high strength gives you noise the model never finished resolving.
    • For a targeted change, masking the region beats raising the strength.

    The mistake to avoid

    Raising strength because the model is not making a change you asked for. If the prompt term is being ignored, more freedom just produces a different picture that also ignores it.

    Related terms

    The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.