Image SafeSearch Classifier

Classify images with an LLM using SafeSearch-style safety categories. Run it on Apify to moderate image URLs or base64 payloads at scale.

What it does

For each image URL (or base64 payload), the Actor uses an LLM to analyze visual content and returns:

Category	Meaning
adult	Adult / explicit content
racy	Suggestive but not necessarily explicit
violence	Violent content
medical	Medical imagery
spoof	Spoof / manipulated content

Likelihood values: UNKNOWN, VERY_UNLIKELY, UNLIKELY, POSSIBLE, LIKELY, VERY_LIKELY.

Derived labels

Label	Rule (defaults)
NSFW	adult or racy ≥ threshold (LIKELY)
Adult	adult ≥ threshold
Violence	violence ≥ threshold
Racy	racy ≥ threshold
Medical	medical ≥ threshold
Spoof	spoof ≥ threshold
Safe	adult / violence / racy / medical all below threshold
Child Safe	adult / racy / violence ≤ UNLIKELY (stricter)

Each dataset item also includes primaryLabel, isNsfw, isSafe, isChildSafe, and reason.

Pricing

This Actor uses pay per result:

• $0.01 per successfully classified image ($10.00 / 1,000 results)
• You only pay for successful classifications written to the default dataset
• Failed images are not charged (they are logged separately)

How to use

1. Open this Actor on Apify
2. Paste one or more public image URLs (or provide base64 in images)
3. Optionally adjust thresholds and concurrency
4. Click Start
5. Review results in the Dataset / Output tab

Input

Field	Required	Description
imageUrls	one of urls/images	Public image URLs
images	one of urls/images	Objects: { id?, url?, base64? }
likelihoodThreshold	no	POSSIBLE | LIKELY | VERY_LIKELY (default LIKELY)
childSafeMaxLikelihood	no	VERY_UNLIKELY | UNLIKELY (default UNLIKELY)
maxConcurrency	no	How many images to analyze in parallel (default 5)

Example input

{
  "imageUrls": [
    "https://example.com/photo-1.jpg",
    "https://example.com/photo-2.jpg"
  ],
  "likelihoodThreshold": "LIKELY",
  "childSafeMaxLikelihood": "UNLIKELY",
  "maxConcurrency": 5
}

Example dataset item

{
  "imageId": "url-1",
  "imageUrl": "https://example.com/photo-1.jpg",
  "reason": "Ordinary photo with no concerning content.",
  "safeSearch": {
    "adult": "VERY_UNLIKELY",
    "spoof": "UNLIKELY",
    "medical": "UNLIKELY",
    "violence": "VERY_UNLIKELY",
    "racy": "VERY_UNLIKELY"
  },
  "flags": {
    "adult": false,
    "violence": false,
    "racy": false,
    "medical": false,
    "spoof": false
  },
  "labels": ["Safe", "Child Safe"],
  "primaryLabel": "Child Safe",
  "isNsfw": false,
  "isSafe": true,
  "isChildSafe": true,
  "error": null
}

Notes

• Image URLs must be publicly reachable so the Actor can download them for analysis.
• For private images, pass base64 in the images array instead of a URL.
• This Actor moderates content; it does not generate or request illegal material.
• LLM inference cost is covered by the Actor configuration; Apify run compute is billed by Apify.