AI Image Detector

Who this guide is for: Anyone running this actor on Apify — via the Console, API, or integrations.
(Developer setup, deploy, and secrets are in the GitHub repo README, not here.)

---

Quick start

AI image detection from URLs — test any number of images in batches of up to 32 URLs per run (start multiple runs or loop the API for larger jobs). Paste public image links, click Start, open the Dataset for scores. Detect AI-generated, deepfake, and generator-specific probabilities (Midjourney, DALL·E, Stable Diffusion, and more). $0.50 per 100 successful detections — failed URLs are not charged.

{
  "imageUrls": ["https://picsum.photos/id/237/400/300"],
  "maxConcurrency": 3
}

---

What this actor does

Provide public image URLs (up to 32 per run — run again for more). For each URL you get one row in the Dataset with:

• Success: probability scores for AI-generated content, deepfake, and specific AI generators
• Failure: a clear error code (e.g. broken link, file too large)

You do not upload files directly — only http/https links to images that are reachable from the internet.

---

Before you run

Your URLs must be:

Requirement	Detail
Public	No login, no signed URLs that expire mid-run
Direct image links	JPEG, PNG, WebP, GIF, etc. — not an HTML page
Size	Up to 8 MB per image
Count	Up to 32 URLs per run — any total volume via multiple runs or API loops
Scheme	http or https only

Pricing: $0.50 per 100 successful detections ($0.005 per image). You are only charged when detection succeeds — see Billing.

Not supported: localhost, private IPs, file://, or URLs behind your firewall.

---

How to run (Apify Console)

Step 1 — Open the actor

1. Go to Apify Console
2. Open Actors → AI Image Detector

Step 2 — Enter input

1. Click the Input tab
2. Paste or edit JSON. Minimal example:

{
  "imageUrls": [
    "https://picsum.photos/id/237/400/300"
  ]
}

3. Or use the Image URLs field in the form UI — one URL per line (up to 32 per run)

Optional fields (defaults are fine for most runs). You cannot exceed the hard limits below — the actor clamps or rejects values server-side, even via API:

{
  "imageUrls": [
    "https://picsum.photos/id/237/400/300"
  ],
  "maxConcurrency": 3,
  "maxRetries": 3,
  "requestTimeoutSeconds": 30
}

If you send	Actor uses
maxConcurrency: 99	5 (capped)
maxRetries: 10	3 (capped)
requestTimeoutSeconds: 600	120 (capped)
33+ URLs	Run fails immediately (not started)

Step 3 — Start the run

1. Click Start (or Save & Start)
2. You are taken to the Run page — wait until status is Succeeded

Typical timing: a few seconds for one small image; longer for many or large images.

Step 4 — Read results

1. On the run page, open the Dataset tab
2. Each row = one input URL
3. Click a row to expand JSON, or use Export → JSON / CSV / Excel

Success: open result → type → ai_generated (0.0–1.0). Higher = more likely AI-generated.

Failure: top-level error field explains what went wrong.

Step 5 — Re-run or schedule (optional)

• Run again: change input → Start
• Schedule: Schedules tab → create a cron job with the same input JSON
• Integrate: use the API tab on the actor page for code snippets and your actor ID

---

Input reference

Field	Required	Default	Description
imageUrls	Yes	—	List of public image URLs (1–32 per run)
maxConcurrency	No	3	Parallel images (1–5, hard cap)
maxRetries	No	3	Retries on transient errors (1–3, hard cap)
requestTimeoutSeconds	No	30	Per-request timeout in seconds (5–120, hard cap)

Values above the maximum (or below the minimum) are automatically reduced to the nearest allowed value. More than 32 URLs rejects the entire run before processing starts.

Large images are automatically resized before analysis — you do not need to compress them yourself.

---

How users access this actor (Console vs API)

This actor runs on Apify’s cloud. Users do not run Python on their machine — they start a run with JSON input and read results from the dataset.

Apify Console (click UI)

1. Paste input → Start
2. Wait for run to finish
3. Open Dataset tab → view or export JSON/CSV

Best for manual tests and small batches.

Apify API (programmatic — “call and get results”)

Typical flow:

POST  /runs          →  start job with { "imageUrls": [...] }
GET   /runs/{id}     →  poll until status = SUCCEEDED
GET   /datasets/{id}/items  →  JSON array of detection results

Example — start and wait in one call (waitForFinish waits up to 120 s):

curl -X POST \
  "https://api.apify.com/v2/acts/YOUR_USERNAME~ai-image-detector/runs?token=YOUR_TOKEN&waitForFinish=120" \
  -H "Content-Type: application/json" \
  -d '{"imageUrls": ["https://picsum.photos/id/237/400/300"]}'

The response includes defaultDatasetId. Fetch rows:

$curl "https://api.apify.com/v2/datasets/DATASET_ID/items?token=YOUR_TOKEN"

Important: This is an async job, not a single instant HTTP response with scores. Each run takes seconds to minutes depending on image count and size. For long batches, poll the run status or use Apify webhooks when the run finishes.

Your actor ID format: YOUR_USERNAME~ai-image-detector (e.g. playful_ultrasound~ai-image-detector). Find exact URLs on the actor page → API tab.

---

Understanding the output

Every dataset row has the same shape:

Field	Meaning
imageUrl	The URL you submitted
mode	upload = processed · rejected = blocked before download (e.g. SSRF)
attempts	How many tries were made (includes retries)
httpStatus	HTTP status from the detection service (null if not reached)
result	Detection scores on success (status + type), or error details on failure
error	Present only when failed — see error codes below

Successful detection

{
  "imageUrl": "https://picsum.photos/id/237/400/300",
  "mode": "upload",
  "attempts": 1,
  "httpStatus": 200,
  "result": {
    "status": "success",
    "type": {
      "ai_generated": 0.02,
      "ai_generators": {
        "midjourney": 0.01,
        "dalle": 0.01,
        "stable_diffusion": 0.01
      },
      "deepfake": 0.01
    }
  }
}

How to interpret scores

Field	Meaning
result.type.ai_generated	Overall AI-generated probability (0–1)
result.type.deepfake	Deepfake probability (0–1)
result.type.ai_generators.*	Per-generator hints (Midjourney, DALL·E, etc.)

There is no fixed “threshold” in the actor — your app decides what counts as “AI” (e.g. ai_generated > 0.5).

Failed detection

{
  "imageUrl": "https://example.com/not-an-image.html",
  "mode": "upload",
  "attempts": 1,
  "httpStatus": null,
  "result": {
    "status": "error",
    "error": {
      "type": "download_failed",
      "message": "download_failed"
    }
  },
  "error": "download_failed"
}

One bad URL does not stop other URLs in the same run — each row is independent.

---

Batch run (multiple images)

{
  "imageUrls": [
    "https://picsum.photos/id/237/400/300",
    "https://picsum.photos/id/238/400/300",
    "https://picsum.photos/id/239/400/300"
  ],
  "maxConcurrency": 3
}

• Up to 32 URLs per run — process any total volume by starting multiple runs or looping the API
• Use Export on the dataset to download all results at once

---

Limits (hard — enforced in code)

These cannot be overridden by input, API, or Console. The actor enforces them on every run.

Limit	Maximum	If exceeded
URLs per run	32	Run fails before processing
maxConcurrency	5	Silently capped to 5
maxRetries	3	Silently capped to 3
requestTimeoutSeconds	120	Silently capped to 120
requestTimeoutSeconds (min)	5	Silently raised to 5
Image download size	8 MB	Row error image_too_large
URL schemes	http, https only	Row error blocked_url / invalid_url_scheme

---

Error codes

Code	What happened	What to do
blocked_url	URL blocked (localhost, private IP, etc.)	Use a public internet URL
invalid_url	Malformed URL	Fix the URL string
invalid_url_scheme	Not http/https	Use https://...
image_too_large	File > 8 MB	Host a smaller copy or resize before linking
download_failed	Could not fetch URL	Check link works in a browser; fix 404/auth/hotlink blocking
decode_failed	Response is not a valid image	Link must point to image bytes, not HTML
compress_failed	Could not prepare image	Try another format (JPEG/PNG)
rate_limited	Too many requests	Lower maxConcurrency, wait, retry
request_failed	Temporary network/service error	Retry the run
service_unavailable	Service in maintenance	Try again later (message suggests ~12 hours)

---

Billing

This actor uses pay per event. You pay for successful AI detections only.

Price

Rate	$0.005 per successful detection
In practice	$0.50 per 100 images (100 successes)
Event name	detection-result

What you pay for

Result	Charged?
result.status = "success"	Yes — one detection-result event
Failed row (bad URL, >8 MB, blocked, rate limit, etc.)	No — row still appears in the dataset

Example costs

Run	Successful images	You pay
1 test URL	1	$0.005
Batch	10	$0.05
Full batch (max)	32	$0.16
Large job	100 (e.g. 4 runs × 25)	$0.50

Failed URLs in the same run do not increase the detection charge (you still pay normal Apify compute for the run itself — see Cost on the run detail page).

Before you start

Set a max cost per run in the Console if you want a spending cap. With 32 URLs, the maximum detection charge is 32 × $0.005 = $0.16 (plus Apify platform usage for the run).

---

Common questions

Can I upload a file from my computer?
No. Host the image somewhere public (S3, R2, CDN, your server) and pass the URL.

Why did my CDN URL fail?
Some CDNs block automated downloads. Test the URL in an incognito window; ensure it returns image bytes, not a login page.

Can I process more than 32 images?
Start multiple runs with 32 URLs each, or call the API in a loop from your own workflow.

Where is my API token?
Console → Settings → Integrations → Personal API tokens.

---

Need help?

Open a run → Log tab for technical details. For billing or access issues, contact the actor owner through Apify.