Media
3 items
Description
Identifies and subtly highlights AI-generated text and images on websites and social networks. eQuantic AI Expose discreetly identifies and flags content created by Artificial Intelligence (language models and image generators) as you browse the web and social feeds. All analysis happens on your device. 🔍 WHAT IT DOES: ✦ Text analysis: • Invisible markers: zero-width characters and other invisible Unicode markers with no job in the text. Composed emoji and writing systems that depend on those characters, such as Persian, Hindi and Khmer, are not accused. • Assistant leftovers: phrases that only appear because an assistant wrote them. • Signature vocabulary: phrasing over-represented in AI text, in Portuguese and English, with the language detected automatically. • Rhythm and format: homogeneous sentence cadence, lexical diversity, and the typical shape of a post generated for professional feeds. These style signals never mark a passage on their own. • Account script: on a LinkedIn profile's activity page, the popup shows whether the account's comments keep repeating the same openers, the habit of an account that follows a script, without marking any comment because of it. ✦ Image analysis: • Content Credentials (C2PA): provenance manifests carrying the standardised digitalSourceType = trainedAlgorithmicMedia assertion, used by tools such as OpenAI DALL-E 3, Adobe Firefly and Microsoft Designer. • Diffusion parameters: prompts, seeds and samplers left in PNG files by Stable Diffusion, ComfyUI, WebUI, NovelAI and InvokeAI. • EXIF / XMP metadata: creator-tool fields pointing at known synthetic generators. • Visible Gemini mark: the glyph Gemini stamps in the corner of an image, which survives when a social network strips the metadata. ✦ Unobtrusive and explained: • Three levels of confidence, each with its own underline: confirmed (solid), probable (dashed) and possible (dotted). Images get a small corner badge. • Hover a highlight to see its level and the exact signals behind it. • The toolbar icon shows how many highlights the page has, and turns grey where the extension is off. • In the popup: the page's result, the master switch, the sensitivity (Low, Balanced or High), text and image analysis, and a button that turns the extension off on the current site only. Everything applies instantly, with no page reload. • Works on any site. On the major social feeds and article sites, adapters pick the right unit of analysis (post, comment or article) and skip interface chrome. • Interface in English, Brazilian Portuguese and European Portuguese. 🔒 PRIVACY: All analysis runs locally on your device. No text you read, image you view, or browsing history is sent to an eQuantic server, and there is no telemetry and no backend. The only traffic the extension generates re-reads images the page already shows. It reads the first 128 KB of an image to parse its provenance metadata. When the metadata settles nothing and the image has the exact size of a generator's native output (Gemini, DALL-E 3 or SDXL presets, or a 512 to 4096 px square), it reads the whole image, up to 24 MB, to look for Gemini's visible mark. Both reads use the browser's cache, so they normally generate no new traffic, and neither carries credentials or any of your data. ⚠️ LIMITATIONS: AI detection is probabilistic. Statistical style signals (rhythm and vocabulary) are evidence, not proof, and can produce false positives on naturally uniform writing such as news leads and academic abstracts. Text written or edited to avoid the known tells leaves the heuristics little to find. Pixel-domain watermarks such as Google SynthID are not detectable by metadata inspection and are out of scope. A negative result is not proof that content is human-made.