One browser extension for all five: text, images, video, speech and documents. Every detector's vote shown, and an honest "inconclusive" when they can't agree.
Free: 1 full English text check a day, 3 document checks, 1 image a month · Video and voice need a free account
Detection Capabilities
Most detectors check text. Ours checks every kind of content you actually see online — in one place.
Checks writing for the patterns left by GPT, Claude and Gemini, scored by each detector separately.
LiveChecks images from DALL·E, Midjourney, Stable Diffusion and Flux for generation and manipulation signals, and reads Content Credentials where a file carries them.
LiveFrame-by-frame analysis plus the audio track. Some generators (including Amazon Nova) can still read as human.
LiveChecks clips for cloned and synthetic voices such as ElevenLabs and OpenAI TTS. Short, phone, or compressed audio is weaker.
LiveUpload a PDF and we inspect the file for tampering and re-save signals. That is not the same as saying an invoice, ID or certificate is authentic or safe to pay.
LiveWhy AI Detector
Most detectors do one thing. We check text, images, video, speech and documents, show you every detector's individual vote, and tell you plainly when they don't agree.
| What you can check | Supported | What we look at |
|---|---|---|
| Text | Yes | Writing patterns scored by each available detector, with every vote shown |
| Images | Yes | Generation and manipulation signals, plus Content Credentials where present |
| Video | Yes | Frame analysis and deepfake signals, plus the audio track |
| Speech | Yes | Synthetic and cloned-voice signals |
| Documents | Yes | Tampering and re-save forensics, which is a different question from authorship |
| Browser extension ✦ | Yes | Check what's in front of you without leaving the page |
We don't publish an accuracy percentage, because no honest detector can. Here's what independent testing actually shows, and how to read a result without over-trusting it.
In your feed
Open a single article or post and, on Pro, run a page scan. Mixed feeds like X home are too large — we refuse those instead of spending detectors on a dump of posts. Image scores can still appear as you browse.
Documents
Upload the file in Document mode. We look at compression, metadata and structure for signs it was edited after the fact. A clean scan means we found no detectable tampering — it is not proof the document is real, and it is not payment authorisation.
Images everywhere
Image results, marketplaces, dating profiles, news photos — the extension scores images for AI-generation signals. That is a probability, not a catch-all.
How It Works
Four steps to a score you can inspect — a probability, not a proof. No PhD required.
Paste text, drop in an image, upload a video, voice note or PDF. Whole-page scan is on Pro. Documents are an upload, not an on-page scan.
Independent detectors analyze writing patterns, pixel artifacts, frame consistency, and voice waveforms in parallel.
Each detector's score is weighted and combined. Where they disagree, or too few could be reached to compare, the result is reported as inconclusive rather than resolved in favour of one of them.
A confidence score with a per-detector breakdown, so you can see exactly which models voted and how sure each one was.
Pricing
Voice and video are expensive to analyse. Plans include a monthly video/voice budget, then prepaid extra scans. Start free, upgrade when you need more.
Perfect for personal use
For individuals & small teams
For teams who see AI content in every format
For organizations
A scan is sent to every detector we can reach for that kind of content, and you see each one's vote, so a single model's mistake doesn't decide your result on its own. How many answer varies — detectors go down, hit rate limits, or decline to score something — and when they contradict each other, or too few respond to compare, we say the result is inconclusive instead of guessing — we would rather tell you we don't know than give you a confident answer we can't stand behind. We've written up what independent testing actually shows about detector accuracy, including where ours is weakest, and how to read a result without over-trusting it.
It looks for text from GPT, Claude and Gemini; images from DALL·E, Midjourney, Stable Diffusion and Flux; video from Sora, Runway and Kling; synthetic voices from ElevenLabs-style cloning tools; and file-level tamper signals in uploaded PDFs. Detectors are tuned per model family, so performance is not uniform across that list — newer generators are generally harder to spot than established ones, and a detector that handles one model well may miss another. That is part of why a result shows each detector's vote instead of one number. Document checks answer a different question again: whether the file looks edited, not whether the paperwork is authentic.
If you have the original file, use that. WhatsApp, Messenger and screenshots usually strip Content Credentials — the file’s own “made with AI” tag — and they wipe pixel traces detectors use. A low AI score on those copies is not proof the picture was a camera photo. We say on the result when the file has no origin tag. On an original ChatGPT or Gemini file you can also try Google SynthID — we do not run that ourselves, it only covers some generators, and a miss does not mean the photo is real.
Only the content you explicitly scan is analyzed. We don't track your browsing history or sell data. See our Privacy Policy.
Paste or upload from anywhere in the popup. Image scores can appear as you browse. Full page scan (Pro) works on a single article or post, not on mixed feeds like X home.
Yes — Pro and Enterprise plans include API access for bulk and automated analysis. Enterprise can also send each completed scan to an HTTPS webhook from the account page. See the API docs.
Starter includes 10 video/voice checks a month, Pro 40, Enterprise 150. If you run out, buy extra scans from your account — 20 for $7, 50 for $15, or 100 for $25. Headline text and image checks stay inside the subscription. The paid detectors (Hive, SightEngine, Sapling) are daily-capped so a worst-case mix still covers their cost; Gemini keeps scoring after that. Extra video/voice scans are prepaid so we never run a detector we have not been paid for.
Yes. Free is 1 device, Starter 2, Pro 3. Enterprise is 5 seats, and each person can stay signed in on 5 computers. Extra logins past that limit sign the oldest one out. Team seats share the same daily and monthly quota — they do not multiply video or voice checks.
Plans, extra scans and billing are on the account page. In the extension, click the Nendry icon, then Account page at the top — or open nendry.com/account. Email support@nendry.com and we reply within one business day.
Yes. Text, images and documents need no account. Video and voice need a free signup. Free includes 1 English text check a day, 1 image or other-language check a month, and 3 document checks a day. Video and voice: 1 a month after you sign in. If a free check is used up, we ask you to get Starter instead of giving a weaker result. Paid plans raise those limits, and Pro adds API access and priority processing.
Try it free in your browser, or add the extension to Chrome — first verdict in under a minute.
On the Chrome Web Store · free to try. Install help →
AI Detector provides probability, not certainty. Detection models can be wrong — that's why we show per-detector breakdowns instead of a single opaque number, and why scores are estimates, not proof. Document mode inspects the file you uploaded for tampering; it does not authenticate identity documents, invoices or certificates, and a clean result is not payment authorisation. For high-stakes decisions, verify with professional fact-checkers.