Image sharpness

The SUPERWISE® Sentinel image sharpness guardrail measures how sharp each image in a prompt is and blocks the request when an image is too blurry. Like image resolution, it works on images, not text — it inspects the images attached to a multimodal prompt before they reach the model.

Use it to keep low-quality, out-of-focus images out of vision requests, where a blurry image wastes tokens and degrades the model's answer.

Action: This guardrail only blocks — there's no content to redact, so there's no remediate option. When an image is below your sharpness threshold, the request is rejected with the Error message you set and never reaches the model. See Manage guardrails.

Parameters

• Minimum sharpness — the lowest sharpness score an image may have and still pass. Sentinel scores each image's sharpness (higher means crisper, lower means blurrier) and blocks any image that scores below this value. Raise it to demand sharper images; lower it to allow softer ones.

An image the gateway can't decode is also blocked.

Example interception

With a minimum sharpness set, a crisp image passes and a blurry one is blocked:

Result: the blurry image scores below the minimum sharpness, so Sentinel blocks the request and returns your error message. The image never reaches the model.

Get Sentinel: Sentinel on superwise.ai.

See also

• Image resolution: block images above or below a resolution limit.
• Manage guardrails: add the guardrail and test it before publishing.
• Guardrails: what each guardrail catches.