This is the complete reference for every image rule type in Cloudinary Moderation: what each one checks, when to use it, and what every setting means.
All rules share the common settings described in "Create a rule" - a name, a description, a Review threshold (how much lands in Needs review versus being decided automatically), and an optional Rejection reason.
This article covers the settings specific to each rule type.
Category | Rule Types |
Background color Brand element placement Centered product | |
AI content detection | |
AI-generated image detection Custom AI-generated image detection Originality check Watermark detection | |
Aspect Ratio Image Quality Image Size | |
Text detection Well-known logo detection |
Composition and background
🎨 Background Color
What it checks: that the background is clean, uncluttered, and matches the color you require - white by default, or any custom color.
Settings:
Background color
Pure white (#FFFFFF) for strict e-commerce standards
White-ish to accept off-white and very light backgrounds.
Custom color to require a specific background color.
Custom background color - the exact background color to require, as a hex value.
Color tolerance - how far a background pixel may deviate from the custom color (0 = exact match only, 0.05 = slightly noticeable differences allowed - the default, 1 = any color).
Allowed off-color area (%) - the maximum share of background pixels allowed to be non-white. A little tolerance forgives soft shadows under the product.
Transparent backgrounds - what to do with images that have a transparent background: Approve, Reject, or Send to review (the default).
When to use it: product photography standards, marketplace compliance.
©️ Brand Element Placement
What it checks: the position, size, and spacing of well-known logos or promotional text in the image - so branded assets follow your layout guidelines.
Settings:
What element to detect? - Brand Logos or Promo texts (detects the largest text block, excluding text on logos)
Position - where the element must appear in the image.
Size (height) - set the Expected height (%), the element's expected height as a percentage of the image height.
Spacing - the clean space above and below the element:
When specifying exact spacing, set the Padding zone size - how much clean space to check above and below, as a percentage of the detected element's height (100% = same height as the element, 50% = half).Logos to detect - when detecting a logo, limit the check to specific brands (type a name and press Enter). Leave empty to detect any logo.
If no brand logo/promo text is detected, the rule should pass - on by default.
When on, images without and logos or promo texts simply pass; t
Turn it off to fail images where the element is missing (use this when the logo or promo text is mandatory).
When to use it: branded templates, campaign assets, and co-branded imagery where your guidelines specify exactly where and how big the logo or promotional text should be.
🪑 Centered Product
What it checks: that the main object is centered in the frame and fills enough of the image to be clearly visible.
Settings:
Maximum Offset - how far the object may sit from the center of the image (0–50%). Lower is stricter.
Minimum Object Coverage - the minimum share of the image the object must fill. Raise it if products look lost in space.
When to use it: product detail pages and catalogs, where consistency across the grid matters.
Custom with AI
🪄 AI Content Detection
What it checks: anything you can phrase as a yes/no question about the image. This is the most flexible rule type - the "Custom with AI" option.
Settings:
Question - a yes/no question that determines whether the image meets your criteria. For example: "Does this image show a person wearing a hat?" Use Generate AI suggestion to help phrase it well.
What does this description represent? - choose Allowed content (images matching the question pass) or Content to block (images matching the question fail).
Tips for good questions: be specific and visual ("Is there a visible price tag on the product?") rather than abstract ("Is this image on-brand?"). One question per rule - create several rules for several criteria.
When to use it: brand guidelines that no built-in rule covers - "no alcohol visible," "must show the product in use," "no competitor packaging in frame."
Originality and authenticity
🤖 AI-Generated Image Detection
What it checks: whether the image is likely AI-generated.
Settings:
Use additional authenticity signals - on by default. Combines the visual analysis with non-visual signals (such as the image's metadata) for higher confidence. Turn it off to judge by visual analysis alone.
Good to know: detection is probabilistic. Borderline images land in Needs review based on your Review threshold, so a human makes the final call where it matters.
When to use it: authenticity standards - product photography that must be real, editorial content, marketplaces, or any place where AI-generated imagery isn't acceptable (or must be labeled).
🤖 Custom AI-Generated Image Detection
What it checks: whether the image is likely AI-generated. Designed for specialized use cases that require tailored detection.
Settings:
Use additional authenticity signals - on by default. Combines the visual analysis with non-visual signals (such as the image's metadata) for higher confidence. Turn it off to judge by visual analysis alone.
When to use it: For specialized use cases that require tailored detection.
🌐 Originality Check
What it checks: whether the image already exists on the public web - helping you avoid copyright issues and unlicensed stock imagery.
Settings:
Where to search - Web sources (the default) searches the public web broadly; Stock-photo sites only narrows the check to stock-photography sites, for catching unlicensed stock imagery specifically.
Ignored sites - sites to skip when matching. Enter a site name or domain (common domain variations are included automatically), add your own domains and your retailers' or partners' sites so legitimate appearances of your images don't count as matches.
Good to know: a match means the image was found somewhere online - it doesn't say whether you have the rights to it. Review matches before acting; licensed stock photos will legitimately appear on the stock site.
When to use it: vetting agency deliveries, user submissions, or any imagery whose provenance you can't fully trust.
💧 Watermark Detection
What it checks: whether visible watermarks appear in the image.
Settings:
Minimum confidence score (1–5) - how confident the detection must be before the rule reacts (default 3). Lower values are more sensitive and catch fainter watermarks; higher values react only to clear, unmistakable watermarks.
Good to know: faint or partially cropped watermarks are easy to miss by eye - zoom in before overriding a rejection.
When to use it: vetting stock imagery, agency deliveries, and user submissions - a visible watermark usually means the image isn't licensed for use.
Quality and size
🖼️ Aspect Ratio
What it checks: whether the image matches an allowed width-to-height ratio.
Settings:
Allowed aspect ratios - Square (1:1), Landscape (4:3, 16:9, 3:2), Portrait (3:4, 9:16, 2:3).
Tolerance Percent - how much the image's ratio may differ from an allowed ratio and still pass. Leave at 0 for exact matches; add a small tolerance to accept near-misses from cropping.
When to use it: when layouts break with the wrong shape - square product grids, story formats, banner slots.
💎 Image Quality
What it checks: overall image quality - resolution, sharpness, and other quality issues.
Settings:
Minimum display size (px) - the size at which the image is evaluated.
Smaller images are resized to this size before being judged, so this should roughly match the size at which your images are actually displayed.
If you show large hero images, raise it; a photo that looks fine as a thumbnail can fail at hero size.
When to use it: as a baseline in almost any image policy - it catches blurry, pixelated, and low-resolution content.
📏 Image Size
What it checks: whether the image meets minimum resolution requirements (width × height).
Settings:
Min Width (px) - required (default 500).
Min Height (px) - required (default 500).
Good to know: unlike other rules, Image Size is a straightforward measurement - an image either meets the minimum or it doesn't - so this rule has no Review threshold. Results are always a clear pass or fail.
When to use it: enforcing platform or marketplace minimums (for example, 1000×1000 for product listings).
Text and logos
🔠 Text Detection
What it checks: whether text appears in the image.
Settings:
Allow text on objects - on by default. Text on the object itself (product labels, packaging, prints) is allowed; only overlaid or background text triggers the rule. Turn it off to flag any text at all.
When to use it: keeping marketing overlays, promo stickers, and burned-in captions out of clean product imagery.
©️ Well-Known Logo Detection
What it checks: the presence of well-known brand logos in the image.
Settings:
Pass if logo is detected - off by default (meaning: pass when no logo is found, fail when one is). Turn it on to require a logo instead.
Brand names - limit the check to specific brands (type a name and press Enter; not case-sensitive). Leave empty to detect any logo.
When to use it: either direction - keeping competitor or third-party logos out of your imagery, or making sure your own logo is present.
Things to review manually
AI-based checks are good, not perfect. Plan for human review where stakes are high:
Originality and logo matches can have lookalikes - verify before acting on a rejection.
Stylized or artistic photography can trip quality and background checks; tune tolerances rather than fighting individual results.
The Review threshold on each rule is your main lever: stricter thresholds send more to humans, looser ones decide more automatically.