Content Safety API - AI Image Moderation & NSFW Detection: Pricing, Documentation

Authentication

All requests require an API key passed via header.

Header	Type	Required	Description
Ocp-Apim-Subscription-Key	string	Yes	Your API subscription key

Request Parameters

Parameter	Type	Required	Default	Description
image_url	string	Conditional	—	Public http(s) URL of the image to analyze. Must return an image/* content-type and be ≤ 4 MB (50×50 to 2048×2048 px). Provide either image_url or image_b64.
image_b64	string	Conditional	—	Base64-encoded image bytes, as an alternative to image_url (same size limits). Provide either image_url or image_b64.
categories	array<string>	No	all four	Restrict analysis to a subset of ["Hate", "SelfHarm", "Sexual", "Violence"]. Omit to analyze all four.
severity_threshold	integer	No	2	Minimum severity (0, 2, 4, 6) counted as flagged. 2 = Low and up (strict), 4 = Medium/High only, 6 = High only. Clamped to 0–6.

Severity levels returned per category: 0 Safe, 2 Low, 4 Medium, 6 High.

Response Fields

Field	Type	Description
flagged	boolean	true = at least one category's severity is ≥ severity_threshold, so the image should be blocked or sent for review. false = every category is below the threshold (safe to allow). Use the per-category severity values to decide how to handle a flagged image.
categories	array	One entry per analyzed harm category.
categories[].category	string	The harm category: Hate, SelfHarm, Sexual, or Violence.
categories[].severity	integer	How strongly that category was detected — 0, 2, 4, or 6 (see Severity Levels below).
severity_threshold	integer	The threshold applied to this request (minimum severity counted as flagged). Defaults to 2; override per request.
ms	integer	Server-side processing time, in milliseconds.

What Each Category Detects

Every image is analyzed against these four harm categories.

Category	Covers
Sexual	Nudity & pornography, sexual acts and imagery, vulgar content, prostitution, abuse, and child exploitation/grooming (including forced or assault sexual content).
Violence	Physical actions intended to hurt, injure, or kill; weapons & guns; gore, injury, and death; bullying & intimidation; domestic violence & physical abuse; terrorism; stalking.
Hate	Attacks or discriminatory content targeting identity attributes — race, ethnicity, nationality, gender identity/expression, sexual orientation, religion, appearance/body size, disability — plus harassment, bullying, and hate symbols.
SelfHarm	Self-injury, suicide, and related acts; eating disorders; and bullying/intimidation that promotes self-harm.

Severity Levels

Each category returns one of four severity levels. flagged is true when any category's severity is at or above severity_threshold.

Severity	Level	Meaning
0	Safe	No objectionable content detected in this category; content is appropriate.
2	Low	Content may be inappropriate in some contexts; low risk.
4	Medium	Content is inappropriate and should generally be filtered.
6	High	Content is highly inappropriate or harmful and should be blocked.

Example: Sexual: 4 with severity_threshold: 2 → flagged: true (4 ≥ 2). The same image with severity_threshold: 6 → flagged: false (4 < 6).

What Each Level Means Per Category (images)

Category	Low (2)	Medium (4)	High (6)
Sexual	Suggestive/racy poses, non-graphic nudity in artwork, body art.	Erotic or soft-nude content, obscured/partial sexual activity.	Explicit sexual acts & organs, CSAM, non-consensual acts.
Violence	Weapons or bladed tools in display or use, light injury, non-realistic violence.	Moderate gore; domestic violence & physical abuse; use of force/coercion to intimidate an individual; a weapon used against a target.	High gore, graphic injury/death, dismemberment, terrorism.
Hate	Stereotyping, biased sentiment, certain hate symbols in context.	Insults, mocking, dehumanization, promotion of exclusion of an identity group.	Hate-speech/hate-crime imagery, glorified extremist/hate symbols.
SelfHarm	Scars, self-harm tools/injury in recovery, emaciation.	Mocking survivors; self-harm shown as a solution; moderate gore.	Explicit self-harm or suicide acts, high gore.

Request Headers

Header	Required	Description
Ocp-Apim-Subscription-Key	Yes	Your API subscription key
Content-Type	Yes	Must be application/json

Response Handling

Status	Meaning
200 OK	Analysis complete. Body contains flagged and per-category severity.
400 Bad Request	Invalid JSON, missing both image_url and image_b64, or invalid categories.
401 Unauthorized	Missing or invalid subscription key.
403 Insufficient Balance	Wallet balance is too low to cover the request.
422 Unprocessable	The image could not be fetched or is not a valid/supported image (bad URL, non-image content-type, > 4 MB, or corrupt).
429 Too Many Requests	Rate limit exceeded. Retry after a short delay.
503 Service Unavailable	Moderation backend temporarily unavailable. Retry.

Limitations & Best Practices

• Strong on graphic / explicit content — gore, weapons, visible injury, and explicit sexual imagery are detected reliably.
• May under-detect implied or staged scenes — a threatening pose or an intimidation scene without visible weapons, gore, or injury can score below the flag threshold even when a human reads it as harmful. Treat a low or zero score as “no graphic violation detected,” not a guarantee that the image is safe.
• Tune the threshold — lower severity_threshold (to 0 or 2) to catch borderline / suggestive content; raise it to flag only clear violations.
• For high-stakes moderation, pair the API with human review of flagged and borderline content.