Detect invisible AI watermarks from ChatGPT, Google Gemini, and Imagen. Works on text, images, and audio — results in under 3 seconds.
1,000 free credits — no credit card required
Every tool below embeds SynthID watermarks — detectable by our system.
Content from Midjourney, Stable Diffusion, Claude, Suno, and others is detected via our analysis-based models — no watermark required.
OpenAI and Google DeepMind announced a landmark partnership: images generated through ChatGPT, DALL·E, Codex, and the OpenAI API now embed SynthID watermarks. This makes SynthID the first watermarking standard adopted across multiple major AI platforms.
An invisible, pixel-level signal embedded directly into every image at generation time. Survives screenshots, JPEG compression, resizing, and cropping — making it significantly more robust than metadata alone.
OpenAI became a C2PA Conforming Generator, attaching signed metadata to content that records its origin, creation method, and publisher. C2PA carries more detail, SynthID carries further — the two layers work together.
What this means for detection
SynthID watermarks now cover two of the largest AI image generators in the world. However, text from ChatGPT does not yet carry SynthID — and content from Midjourney, Stable Diffusion, and most other generators remains watermark-free. That's where our analysis-based detection fills the gap.
SynthID is Google DeepMind's invisible watermarking system, originally built for Google AI products like Gemini and Imagen — and now also adopted by OpenAI for ChatGPT image generation. Unlike a visible stamp or metadata tag, SynthID embeds its signature at the generation level — inside pixel values, audio spectrograms, or token probability distributions. The result is content that looks completely normal to a human but carries a hidden, machine-readable identity.
Pixel-level embedding in Gemini, Imagen, ChatGPT, and DALL·E. Survives JPEG compression, resizing, cropping, and color adjustments.
Token probability shifts during generation create statistical patterns that survive paraphrasing. Currently Google Gemini only.
Frequency modifications in the spectrogram — inaudible to humans but detectable by analysis. Used in Google AI audio tools.
Our detector uses a multi-layered approach — combining watermark pattern analysis with statistical and semantic modeling — so it catches both watermarked and non-watermarked AI content.
Paste text, upload an image, or drop an audio file. Our system accepts all common formats — JPEG, PNG, MP3, WAV, and more.
For images, we run frequency-domain analysis to detect SynthID's pixel-level signal, plus C2PA metadata checks. For text, we analyze token distributions — the exact patterns SynthID and similar systems embed.
Watermark results are cross-checked against our own AI detection models. If SynthID isn't present but content shows AI signatures (from Midjourney, Claude, etc.), we flag those too.
You get a clear result: SynthID watermark detected, AI-generated without watermark, or likely human-written — with a confidence score.
SynthID now covers Google and OpenAI image tools. Our detector combines watermark detection with analysis-based models so you're covered regardless of source. See also: AI Detection Accuracy Comparison.
| Content source | SynthID only | Our detector |
|---|---|---|
| ChatGPT / DALL·E images | ||
| ChatGPT text | ||
| Google Gemini text | ||
| Google Imagen images | ||
| Midjourney / Stable Diffusion | ||
| Claude (Anthropic) | ||
| Survives screenshots & compression | ||
| Works without any watermark |
SynthID is Google DeepMind's invisible watermarking technology. It embeds undetectable signatures into AI-generated images, audio, and text — directly during the generation process — without any visible mark on the content.
Yes — as of May 2026, OpenAI and Google partnered to embed SynthID watermarks into images generated via ChatGPT, DALL·E, Codex, and the OpenAI API. This is in addition to C2PA content credentials that OpenAI has been adding since 2024. Text from ChatGPT does not yet carry SynthID.
Yes. SynthID for text works by subtly adjusting the probability distribution of word choices during generation, creating statistical patterns that our detection tools can analyze. Paste the text into our detector and get a result in seconds.
Yes. Because SynthID is embedded at generation time rather than added as metadata, it survives common edits like JPEG compression, resizing, cropping, color adjustments, and screenshots. C2PA metadata (which OpenAI also adds) can be stripped, but SynthID remains.
C2PA (Coalition for Content Provenance and Authenticity) is an open standard that attaches signed metadata to content, recording how and where it was created. OpenAI became a C2PA Conforming Generator in 2026. SynthID and C2PA are complementary: C2PA carries detailed context, SynthID provides a pixel-level watermark that survives when metadata is stripped.
No — SynthID only covers Google and OpenAI AI tools. For content from Midjourney, Stable Diffusion, Claude, and others, we use our own analysis-based detection models which work across all AI generators regardless of watermark.
Yes. You get 1,000 free credits when you create an account — no credit card needed. Each text detection uses 1 credit per word, so you can run many checks for free.
A deep dive into the technology behind SynthID — pixel embedding, token probability shifts, and audio spectrogram modification.
SynthID, C2PA, and the broader landscape of content provenance standards — what's being adopted and where things are heading.
How watermark-based methods like SynthID compare to analysis-based detectors across accuracy, coverage, and real-world use cases.
Why SynthID and similar watermarking approaches are only part of the solution in the ongoing battle between AI generators and detectors.
Disclaimer: WasItAIGenerated is an independent AI detection service and is not affiliated with, endorsed by, or in any way officially connected to Google, Google DeepMind, OpenAI, YouTube, or any other company mentioned on this page. SynthID is a trademark of Google DeepMind. ChatGPT, DALL·E, Codex, and Sora are trademarks of OpenAI. All other product names, logos, and brands are the property of their respective owners and are used here solely for informational and descriptive purposes. References to third-party products and announcements are based on publicly available information.
Paste text, upload an image or audio file. Covers ChatGPT, Gemini, Imagen and more. Result in under 3 seconds.
Start for Free1,000 free credits — no credit card needed