Skip to content

Latest commit

 

History

History
129 lines (72 loc) · 6.27 KB

File metadata and controls

129 lines (72 loc) · 6.27 KB

TruthLens — Frequently Asked Questions

Short answers to the questions people ask most often about AI detection, deepfakes, and how TruthLens works. Tuned for search — if you arrived here from Google or Bing, the answer you're looking for is likely below.


General

What is TruthLens?

TruthLens is a free, open-source browser extension that flags AI-generated text, deepfake images, and cloned voices on any webpage. It runs entirely on your device — no servers, no accounts, no telemetry — and it ships under the MIT license.

Is it really free? What's the catch?

It is genuinely free. There is no paid tier, no sign-up, no trial, no usage cap. The "catch" is that it's a heuristic tool and not a courtroom-grade forensic detector — which we say loudly throughout the documentation. You get what every paid AI-detector vendor claims to offer, except the source is public so you can verify the claims yourself.

How is TruthLens different from GPTZero, Originality.ai, Copyleaks, or Winston AI?

TruthLens Most paid detectors
Price Free, forever $10–$30 / month and up
Where data goes Your browser only Uploaded to vendor servers
Source code MIT, public Closed-source black box
Browser-native Yes No (web-paste workflow)
Images and audio Yes Often text-only

If you need a single official verdict for a tribunal, use a forensic service. For everyday triage on the open web, TruthLens is faster, free, and more transparent.


Detection capability

What kinds of AI content can it detect?

  • Text generated by ChatGPT, GPT-4, GPT-4o, Claude (Opus / Sonnet / Haiku), Gemini, Llama, Mistral, and other large language models.
  • Images generated by Midjourney, DALL·E 2/3, Stable Diffusion (SD1.5, SDXL, SD3), Flux, Leonardo, Ideogram, Adobe Firefly, Playground, Civitai, Lexica, and Replicate-hosted models.
  • Audio / voice synthesized by ElevenLabs, OpenAI TTS, Resemble, Play.ht, Murf, and similar voice-clone services — including the voice tracks in <video> elements.

Can it detect deepfake videos?

Today it detects the voice track inside a <video> element using the audio detector. Frame-level deepfake video detection is on the public roadmap.

How accurate is it?

Accuracy depends entirely on the source material, and no honest AI detector should give you a single accuracy percentage. TruthLens surfaces the underlying signals with each verdict, so instead of "trust this number", you see why something was flagged. That is the only honest answer in 2026.

Will it produce false positives?

Yes. Any AI detector that says it won't is lying. Confident, formal human writing — corporate prose, government press releases, certain academic styles — can fire the LLM-phrase heuristic. Always read the signal breakdown before treating a verdict as fact.

Will it produce false negatives?

Yes. A skilled human editor can paraphrase LLM output until the statistical fingerprints fade. Adversarial AI text (generated, then re-prompted to "sound human") is the hardest case for every detector on the market — including ours.


Privacy

Does TruthLens upload my data?

No. Open the Network tab of your browser's devtools and watch — when you scan a page, zero outbound requests fire. The only storage is your settings (Chrome sync) and a local counter of how many items you've flagged.

Does TruthLens use my browsing data to train a model?

No. There is no model. The detectors are pure JavaScript heuristics committed to the repo. Read every line.

What permissions does the extension need and why?

Permission Why
storage Persist your settings and the local flag counter
activeTab Inject the scanner into the page you're looking at
contextMenus Right-click → "Scan with TruthLens"
scripting Re-inject the content script when triggered manually
<all_urls> host Lets you scan any page — declared as a host permission, not a content-script auto-injection vector for content reading. The content script only reads visible DOM; nothing is exfiltrated.

Compatibility

Which browsers are supported?

Chrome, Microsoft Edge, Brave, Opera, Arc, Vivaldi — and any other Chromium-based browser that supports Manifest V3.

Firefox? Safari?

On the roadmap. Firefox needs the Manifest V3 polyfill for chrome.scripting. Safari requires Xcode-side packaging.

Will it work on mobile browsers?

Kiwi browser on Android can load Chrome extensions. iOS Safari extensions are a separate port that's planned but not started.


Use cases

Can I use TruthLens to catch students using ChatGPT?

You can use it as a first-pass triage — TruthLens will reliably flag unedited LLM output and obvious phrase fingerprints. You should never use it as the sole evidence for a misconduct decision. Pair it with a conversation, drafts, and version history.

Is it useful for journalists?

Yes — it's one of the use cases the project was built around. When breaking-news photos surface on social media, a quick scan tells you whether the source URL is a known AI host (Midjourney CDN, DALL·E blob storage, Leonardo, etc.) and whether the pixel statistics look natural.

Can I use it commercially?

Yes. MIT license. Use it inside your newsroom, classroom, or company. Forks are encouraged.


Contributing

How do I add a new LLM phrase fingerprint?

Open detectors/text-detector.js, add a regex to LLM_PHRASE_FINGERPRINTS, and open a PR with an example or source.

How do I add support for a new AI image service?

Append the host(s) to KNOWN_AI_HOSTS in detectors/image-detector.js.

Can I sponsor this project?

Sponsorship configuration lives in .github/FUNDING.yml. PRs and contributions are also valuable — many of them more than cash.


Related searches

People also search for: AI detector free, best ChatGPT detector, deepfake detector online, how to spot AI-generated images, how to detect AI writing, Midjourney watermark detection, voice clone detector free, synthetic media identifier, GPTZero alternative, Originality.ai alternative, free deepfake detector, Chrome extension AI detector, open source AI detection tool, on-device AI detector.