Images are the heart of Cloudinary Moderation: This article covers what image moderation checks, how to run it, and how to read image results.

What image moderation can check

Image rules cover five broad areas:

🎨 Composition and background - whether the background is clean and matches the required color (Background Color), whether the main object is centered and prominent (Centered Product), and where logos or promo text sit in the frame (Brand Element Placement).
🪄 Custom with AI - anything you can describe as a yes/no question (AI Content Detection).
🛡️ Originality and authenticity - AI-generated images (AI-Generated Image Detection), images that already exist on the public web (Originality Check), and visible watermarks (Watermark Detection).
💎 Quality and size - allowed shapes (Aspect Ratio), resolution, sharpness, and general quality issues (Image Quality), and minimum dimensions (Image Size).
🔠 Text and logos - text in the image (Text Detection) and well-known brand logos (Well-Known Logo Detection).

See "Image rule types and how to use them" for every rule and setting in detail.

How to moderate images

Select Run Moderation > One-time moderation (or set up ongoing moderation for new uploads).
In Select Assets, keep the asset type on Images, then choose a folder or search for the images you want.
In Select Rules, pick image rules or an image policy.
Confirm and run.

Any mix of image rules can run together in the same run or policy.

Each image appears as a card with its result: ✅ Approved, ❌ Rejected,
or 🚩 Needs review.
Select an image to see its Rule Results - per rule, the outcome, a confidence percentage, and an explanation of what was found ("Watermark detected in bottom-right corner").
Open Preview mode to review images full-size, one at a time, and approve (A), reject (R), or skip (S) with the keyboard.

Needs review first - these are the images the system deliberately left for you.
Borderline quality calls - compression artifacts and intentional stylistic blur can look similar to a quality rule; the confidence score tells you how sure the check was.
Small details - faint watermarks and tiny text are exactly what the rules are good at spotting and humans tend to miss; zoom in before overriding a rejection.