Frequently asked questions

SynthGuard is a privacy-first anti-AI-detection toolkit. Image humanizer, video humanizer, text humanizer, and a free AI image detector — all running in your browser, no uploads.

Because the moment your content touches a server, you've lost control of it. Browser-only means there's no server log, no training corpus contribution, no data breach risk.

3 humanize credits per week, plus 10 AI-detector scans (separate pool). Credits spend at 1 per image, 2 per short video, and 0.5–1.5 per text run (light/standard/aggressive). Pro ($14.99/mo) gives 50 weekly credits and 200 detector scans; Studio ($29/mo) is unlimited under fair use.

Account email, plan tier, weekly credit count, and an event log of when humanizations ran (no content). That's it.

Tested against GPTZero, Originality.ai, Turnitin, Winston AI, Copyleaks, ZeroGPT. Detection landscape changes — always verify.

English and German with native phrase dictionaries. Other languages get the structural layers (restructure, punctuation) but not phrase-replace.

No. All layers are meaning-preserving by design. Phrase-replace uses semantically equivalent substitutions, restructure keeps clauses intact.

There's no hard cap. 50,000+ words processes in seconds because everything runs locally.

Live measurement of sentence-length variance — the metric most AI detectors score. Above ≈0.55 generally reads as human.

Hive Moderation, Sightengine, Illuminarty, and the platform-side classifiers used by Instagram, TikTok, Reddit moderation bots, OnlyFans, Etsy.

At default settings, no visible loss. Strict profile adds slightly more perturbation but stays below human perception threshold.

Re-encodes so C2PA is not carried over, injects PRNU sensor noise, disrupts FFT signatures, generates realistic EXIF, perturbs texture statistics, plus refinement passes. Some layers are intentionally not detailed publicly.

It disrupts the synchronization signal SynthID relies on, which collapses detection to chance. The image looks identical.

Yes — drop multiple files into the humanizer and they process in parallel via Web Workers.

MP4, MOV, WebM. Output is always MP4 with stripped container metadata.

Clips are capped at 30 seconds in the current version to keep in-browser exports fast and reliable. Sources up to 4K are accepted and downscaled for export.

Mild — re-encoding always costs a small amount. Output is calibrated to look the same to viewers.

1 video = 2 credits. Free plan gets 1 video per week (3 credits, video uses 2).

Because verifying your humanized image scores as human is the natural last step of the workflow. Charging for it would defeat the purpose.

It scores on multiple signals (PRNU, FFT, EXIF, texture statistics) and returns a 0–100 confidence. Rigorous on current AI generators, less reliable on edge cases like edited real photos.

No — it has its own weekly pool, separate from humanize credits: 10 scans/week on Free, 200 on Pro, unlimited on Studio.

No. Same architecture as the rest of the suite — all analysis runs in your browser.

Yes. Cancel from your hub account page (/hub?view=account) or use the public /cancel-contract form. No login required for the form (compliant with §312k BGB).

EU statutory withdrawal period applies. After that, prorated refunds for unused time on annual plans only.

ISO Monday 00:00 UTC. So on Monday morning everyone is back to their plan's weekly cap.

Not yet. On the roadmap.