⚠ IMPORTANT INFORMATION
The AI watermark landscape (June 2026): Google's SynthID is now embedded in images from ChatGPT, Codex, the OpenAI API, Gemini, Imagen, Veo, Lyria, NVIDIA Cosmos video, ElevenLabs audio, and Kakao. Other AI tools — FLUX, SDXL, Stable Diffusion, Midjourney, Apple's ImagePlayground, Anthropic's Claude — do not currently use SynthID.
AIGNCY defeats AI detection across every major social and content platform. These platforms detect AI content through C2PA metadata, which AIGNCY strips completely. Many platforms also strip metadata themselves on upload as a side effect of recompression.
| Platform |
How they detect AI |
Defeats |
| Instagram / Facebook / Threads | C2PA metadata | ✓ |
| X (Twitter) | None — strips C2PA on upload | ✓ |
| TikTok | C2PA + creator disclosure | ✓ |
| LinkedIn | C2PA metadata (preserved on upload) | ✓ |
| Reddit / Discord / WhatsApp | None for AI watermarks | ✓ |
| YouTube | Creator self-disclosure only | ✓ |
| Chrome right-click → SynthID Detector | Pixel-level SynthID watermark | ✗ |
One stricter edge case: Google announced at I/O 2026 (May) that Chrome would gain a right-click "is this AI?" check via the SynthID Detector — rolling out in waves, so some users have it now and more will get it over coming months. The Gemini app at gemini.google.com is live for everyone today and runs the same SynthID Detector. This bypasses platform-level detection entirely — the check happens at the pixel level on the individual viewer's device. AIGNCY's pixel-disruption methods significantly weaken SynthID but don't always defeat this direct test.
For users worried about that case: regenerate the image through a non-Google diffusion model first (FLUX.2 Pro Edit via fal.ai or WaveSpeed at ~$0.05/image, or run Stable Diffusion / SDXL img2img locally at ~0.5 denoising strength). Running the image through a different diffusion model physically destroys the SynthID signal. Then process the result through AIGNCY to clean metadata and inject the camera fingerprint.
What each AIGNCY method does:
- Strip metadata — removes embedded C2PA tags, AI tool fingerprints and generator IDs that platforms increasingly scan for.
- Inject phone EXIF — writes realistic iPhone or Samsung Galaxy camera metadata, so the "no EXIF = AI" heuristic sees a normal phone photo.
- Mesh warp — subtle non-uniform geometric displacement that breaks the spatial grid most spread-spectrum watermarks rely on.
- Frequency disruption — custom DCT quantization that aggressively rounds the mid-frequency bands where AI watermarks typically hide.
- Pixel noise (Gaussian), micro-blur, colour jitter, dither, JPEG recompression — together these disrupt the perceptual hashes platforms use to track and identify generated images across re-uploads.
Bottom line: effective across every major social media and content platform. Chrome's individual SynthID checking requires the diffusion-regen workaround above. Detection systems evolve, so will the tool — AIGNCY is updated regularly to stay ahead.
AIGNCY cleans AI-generated images and videos so they don't carry the tell-tale traces that mark them as AI. Everything runs in your browser — your files are never uploaded anywhere.
IMAGES
- Open the app and drag an image in (or click to choose one).
- It automatically strips the hidden metadata — the embedded tags that say where the file came from, what made it, and when.
- It also works to remove AI watermarks like SynthID, the invisible markers some AI image tools bake into the picture.
- (Optional) Use the controls to add iPhone-style camera info, a date, and a location so the photo reads like a real phone shot.
- Click EXPORT to download the cleaned image.
TIP: If you use the noise option in the Synth settings, test it first. Noise comes out stronger on low-resolution images, so it usually takes a few goes to get the level right before you settle on it.
VIDEO
- Click the VIDEO DATA FIX tab at the top (on phones, tap the VIDEO button in the bottom bar).
- Drop in your AI video — Kling, Veo, Sora, Runway, etc. (MP4 or MOV).
- It strips all the AI metadata: the source/provenance tags, the original timestamps, and the encoder fingerprints.
- Pick an iPhone model, a date, and a location (or hit AROUND HERE to jitter the selected city). It writes genuine iPhone camera info and GPS in their place.
- Click FIX & DOWNLOAD. The video is never re-encoded (no quality loss) — it's a clean, lossless copy with new details.
The first video you do downloads the engine (about 30 MB, one time only), then it's instant after that.
