VISIONContent moderation for platforms that can’t afford mistakes.
Detectivision checks text, images, video, audio and live streams in real time, with models we build and train in-house.
…when John, that , came in and started giving advice.
profanity · 0.94
Text, images, video and audio.
Every request returns the flags it raised and how sure the models are. In text and images each flag judges the whole item, so the answer says what was found, not where.
Yesterday, I was working on the new marketing campaign when John, that idiot, came in and started giving me unsolicited advice.
Detect
detectThe flags raised and how sure the models are of each.
Capture
captureThe default. It also lists what each flag found, with the second and the box for flags that say where, such as those of video.
Censor
censorA copy with the findings hidden, from flags that say where. Text and image flags judge the whole item, so for them it answers like detect, at the same price.
Live streams.
Every frame, as it airs.
Give a stream link or send frames from a camera; every frame you choose is checked as it airs, and every flagged moment reaches your webhook with its picture.
- flaggedch-09 · 31:38200 · 102 ms
- flaggedch-03 · 11:57200 · 65 ms
- flaggedch-09 · 31:29200 · 118 ms
- flaggedch-03 · 11:48200 · 81 ms
- flaggedch-09 · 31:20200 · 134 ms
- flaggedch-03 · 11:39200 · 97 ms
Photos: Szabo Viktor, Higor Hanschen, Sven Kucinic, ELLA DON, Illia Horokhovsky / Unsplash
Every frame
fpsThe frames you choose, up to 30 a second, are checked for your flags.
Moments
live.momentEach flagged frame is kept with its picture and sent to your webhook while the stream is still on.
Any stream
rtsp · rtmp · hls · srtA link to a camera or a broadcast, or frames you send from a browser or a device. Paid by the minute.
Not a ChatGPT wrapper. Our own models.
Detectivision runs on NLP, vision and audio models that our R&D team researches, trains with machine learning and develops in-house.
Language
NLP
Text models that find insults, hate speech, profanity and personal data, in 12 languages.
Vision
CV
Image and video models that locate nudity, violence, smoking and alcohol down to the region.
Audio
ASR
Speech models that transcribe and classify audio with second-level timestamps.
Real-time
RT
A streaming pipeline that checks live broadcasts frame by frame as they air.
- 01
Send content or a stream to the API, or try it in the Playground
- 02
The models check it for your flags and say where each finding is
- 03
Your team reviews what was flagged, in the Dashboard or your own tools
- 04
You publish, hide or replace it with the censored copy, and record what you did
Integrate in an afternoon.
Send content, choose a mode and get structured JSON back. Create a key in the Dashboard and make your first call in minutes.
- POST/api/v1/moderationText and images
- POST/api/v1/live/sessionsLive streams
- GET/api/v1/moderation/{id}A result
- GET/api/v1/flagsYour flags
curl https://api.detectivision.ai/api/v1/moderation \
-H "Api-Key: $DETECTIVISION_API_KEY" \
-H "Content-Type: application/json" \
-d '{"type": "text_moderation", "input": "that idiot came in again", "mode": "detect"}'{
"type": "text_moderation",
"mode": "detect",
"status": "completed",
"flagged": true,
"flags": [{ "flag": "insult", "confidence": 0.94 }],
"usage": { "charged": 2, "remaining": 9998 }
}Pay for what you check.
Basic
Free for 2 monthsDiscountedWas$49$29per month- 10,000 text credits
- 10,000 image credits
- 300 video credits
Pro
DiscountedWas$199$99per month- 100,000 text credits
- 100,000 image credits
- 2,000 video credits
- 2,000 audio credits
- 1,000 live stream credits
Enterprise
Custom- Custom AI model training for images and video
- Custom models for live streams
- On-premise deployment on your own hardware
- Volume priced for you, with no fixed quota
- Dedicated support
Questions, answered.
Something else? Write to support@detectivision.ai
No. Detectivision runs on NLP, vision and audio models that our R&D team researches, trains and develops in-house.
In text: insults, hate speech, racism, sexism, sexual content, self-harm and talk of drugs, drinking, smoking or gambling; the Basic plan also flags profanity. In images: sexual content, violence, hate symbols and scams; the scam flag is also raised by contact details, such as a phone number written on the picture. In video: nudity, sexual objects, smoking, alcohol and brand logos, with the second and the box of each finding. In audio: profanity, insults, hate speech, sexual content, self-harm and talk of bad habits, with the seconds they were said. Live streams get the visual flags frame by frame and the flags of speech as it comes in. Your account’s flags come from GET /api/v1/flags.
English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian, Hindi, Chinese and Japanese. Text in other languages is checked too, but the results are not reliable.
Yes, for video and live streams: in censor mode a video comes back as a copy with every finding hidden, and each flagged moment of a live stream gets a censored copy, blurred, pixelated or boxed as you choose. In text and images every flag judges the whole text or image: the answer says which flags were raised and how sure the models are, not where, so no censored copy comes back, whatever mode you ask for. Hide or blur flagged content in your own app.
Yes. Give a stream link (RTSP, RTMP, HLS, HTTP or SRT) or send frames from a camera: every frame you choose is checked as it airs, and each flagged moment reaches your webhook with its picture. A session is paid for the frames checked.
In credits. Each kind of content has its own balance, and a check that cannot finish costs nothing. Basic is free for 2 months, with 10,000 text, 10,000 image and 300 video credits. The plans and every price are on the pricing page.
Create an API key in the Dashboard and call the REST API with it. The Playground shows the cURL, Python and Node.js code of every request, and the docs cover the rest.
Tell us what you need to moderate.
We will get back to you with a demo and details.
support@detectivision.ai