What it actually costs to scan a video for pornographic content, across the providers that publish a price. Every figure below is quoted from the vendor's own pricing page, checked on 4 September 2026, with the source named. Where a number is derived rather than published, it says so.
| Provider | Published video rate | Free allowance | Max file size |
|---|---|---|---|
| SafeReel | €0.020 / min, billed per second, max 20 billed minutes per video | 100 video-minutes/month, ongoing | No limit — videos fetched by URL |
| AWS Rekognition | $0.10 / min (Content Moderation, us-east-1) | 60 min/month, first 12 months only | Video must be stored in S3 |
| Google Video Intelligence | $0.10 / min (explicit content detection) | First 1,000 min/month, ongoing | Video must be stored in GCS |
| Sightengine | $0.002 / operation — ~$0.12/min derived, see below | 2,000 operations/month (max 500/day) | 50 MB (Starter) · 500 MB (Pro) |
| Hive | Not published — "Contact Sales" | $50+ in credits after adding a card | Not published |
| JigsawStack | $0.99 / million tokens, where 1 token = 1 ms of processing | 1M tokens/month | Not published |
Two things move the real bill more than the headline rate.
1. SafeReel caps billing at 20 minutes per video. A 90-minute film is analysed in full but billed as 20 minutes — €0.40. The same file at $0.10/min costs $9.00. That is roughly 21× cheaper on a feature-length video, versus about 4.5× cheaper minute-for-minute in the first 20 minutes. (€0.020 is approximately $0.021–0.023 depending on the day's rate; the comparison is indicative, not a quoted conversion.)
| Video length | SafeReel | At $0.10/min |
|---|---|---|
| 3 minutes | €0.06 | $0.30 |
| 20 minutes | €0.40 | $2.00 |
| 45 minutes | €0.40 (capped) | $4.50 |
| 90 minutes | €0.40 (capped) | $9.00 |
| 3 hours | €0.40 (capped) | $18.00 |
2. File-size limits decide whether the job is possible at all. Sightengine documents a maximum of 50 MB per video on the Starter plan and 500 MB on Pro (source: sightengine.com/pricing, "Compare Plans in Detail"). A 2 GB upload cannot be submitted on either plan at any price. AWS Rekognition and Google Video Intelligence both require the file to already sit in their own object storage — S3 and GCS respectively — so a video hosted anywhere else has to be copied in first, at the cost of the transfer and a second copy of the data.
SafeReel takes a public URL and fetches the file itself. There is no file-size limit, no upload step, and no second copy: the video stays where it already lives, whether that is S3, R2, a CDN, or your own origin.
Sightengine bills operations, not minutes. Their pricing page defines one operation as one model run on one image, and lists a configurable video frame rate of 0.5–2 fps on Starter and Pro. At 1 fps — the midpoint, and the common default — one minute of video is 60 frames, so 60 operations. At the published overage rate of $0.002 per operation that is $0.12 per video-minute.
This figure is a derivation, not a Sightengine quote. At 0.5 fps it halves to $0.06/min with correspondingly coarser sampling; at 2 fps it doubles to $0.24/min. Included-plan operations change the effective rate again. Check your own frame rate and plan before relying on it.
Price is the easy axis. Three others matter more once you are in production:
Published accuracy. Of the providers above, accuracy claims are almost always marketing figures with no methodology attached. SafeReel publishes a full accuracy report: 95.5% accuracy, 92.2% recall and 0.41% false positives measured on 1,089 production videos, with the caveats stated. Ask any vendor for their sample size and their false-positive rate — the second number is the one that annoys your users.
Scope. SafeReel detects pornographic content and nothing else. It makes no age determinations and does not offer CSAM detection. Sightengine and Hive cover far broader class sets — violence, weapons, drugs, hate symbols, self-harm. If you need those, that breadth is worth paying for; if you only need an adult-content verdict, you are paying for classes you will never read.
Retention. SafeReel processes image bytes in memory and deletes video frames within minutes of the verdict, keeping only verdicts, quota counters and non-reversible hashes. See Trust & privacy. Check each vendor's retention terms against your own DPA before you compare rates.
These figures are checked by hand and vendors change pricing without notice. If something here is out of date or wrong, tell us at [email protected] and we will fix it and note the date.
Sources, all retrieved 4 September 2026: aws.amazon.com/rekognition/pricing (Content Moderation, $0.10/min, us-east-1 pricing example) · cloud.google.com/video-intelligence/pricing (explicit content detection, $0.00 for 0–1,000 min, $0.10/min above) · sightengine.com/pricing (plans, $0.002 per additional operation, file-size and frame-rate limits) · thehive.ai/pricing (no published per-unit rates) · jigsawstack.com/pricing (token model) · safereel.ai (own published pricing).