Check an image for AI watermarks
Upload an image and RAIW lists the AI provenance signals it carries: metadata, platform labels and visible marks. The check is free and runs in seconds.
Quick clean is free with Standard output up to 12 MP. Keeping the original resolution above that size requires paid processing.
No account, no subscription
Deletion requests supported
Refund available if paid removal fails
What the checker looks for
- Content Credentials (C2PA) - the signed provenance label added by ChatGPT, Gemini, Adobe Firefly, Microsoft, Stability AI, Black Forest Labs (FLUX), ByteDance (Doubao / Jimeng), Canva, ElevenLabs, Bria AI, and fal.ai; the checker names the signing platform.
- SynthID - the invisible watermark used by Google and OpenAI image tools, detected through the metadata companion that Google and OpenAI tools attach.
- Made with AI label - the IPTC tag that triggers AI disclosures on Instagram, Facebook, and X.
- Generator metadata - EXIF/XMP software tags and embedded Stable Diffusion / ComfyUI generation parameters.
- Visible Gemini sparkle - the on-image logomark, found with on-device image analysis (no GPU).
- China AIGC label (TC260) - the disclosure tag used by Doubao and other China-served generators.
- Forensic and newer-standard tags - a C2PA soft-binding naming a watermark vendor (Adobe, Digimarc, Imatag) and the newer IPTC 2025.1 AI-disclosure fields.
- xAI Grok signature - the EXIF Signature blob and UUID Artist recognized in the current Grok (Aurora) profile. Other or future signals may not be detectable.
- Samsung Galaxy AI edits - the proprietary genAIType marker that Galaxy AI tools (Generative Edit, Sketch to Image) leave in the photo's metadata; the visible on-image Galaxy AI corner label is also detected for the Italian-locale wordmark.
- Apple Photos Clean Up edits - the credit tag Apple Intelligence writes when its Clean Up tool erases an object from a photo; the checker names Apple Photos as the source instead of a generic Made-with-AI tag.
- Hugging Face model origin - file metadata showing the image came from an AI model hosted on Hugging Face.
- Visible Doubao / Jimeng mark - the on-image corner label that ByteDance's Doubao and Jimeng (Dreamina) stamp on their pictures, also found with on-device image analysis (no GPU).
Supported visible marks
The local visible-mark scan covers the Gemini sparkle and labels from Doubao, Jimeng, Qwen, Kling, Tencent Yuanbao, Samsung Galaxy AI, RunningHub, Baidu, and LibLibAI. It runs without a GPU. The Samsung label is calibrated for one Galaxy interface language, so the same mark in another language may not be found.
What the checker can't do
RAIW reads provenance signals; it does not run a pixel-level "is this AI?" classifier, and it never declares an image human-made. A clear signal means the image was almost certainly made or labeled by an AI tool. But the absence of a signal is not proof: metadata is routinely stripped when an image is re-saved, screenshotted, or uploaded to social media, and some watermarks - such as the SynthID pixel layer - can only be confirmed with the tool that made the image. For an authoritative pixel-level check, use the Google Gemini app or openai.com/verify.
How to check an image
- Upload a PNG, JPG, WebP, AVIF, HEIC, or HEIF image (up to 150 MB). No account needed.
- Read the plain-language verdict: what was found and the likely source.
- If a watermark or label is found, switch to a removal mode to clean it.
Related tools
Remove SynthID watermarks
Disrupt the invisible SynthID marker after you detect it.
Remove C2PA Content Credentials
Strip the signed provenance manifest from the file.
Remove the Made with AI label
Clean the IPTC and C2PA metadata that triggers platform disclosures.
Remove the Gemini watermark
Erase the visible sparkle and hidden Gemini signals.
Remove visible and hidden watermarks
Use the All-in-one mode to remove visible, invisible, and metadata signals.
Frequently asked questions
How do I check if an image has an AI watermark?
Upload it to RAIW's checker. It scans for AI provenance signals - Content Credentials (C2PA), SynthID markers, the Made with AI label, generator metadata, the visible Gemini sparkle, and the China AIGC (TC260) label - and shows a plain-language verdict in seconds. No signup required.
Can this tool tell whether any image is AI-generated?
No. It detects provenance signals that AI tools embed, not a pixel-level guess. If signals are present, the image was almost certainly made or labeled by AI. If none are found, the image may still be AI: these signals are often stripped by re-saving, screenshots, or social-media upload, and some (like the SynthID pixel watermark) need the original vendor's tool to confirm.
What is a C2PA Content Credential and how do I check it?
C2PA Content Credentials are a signed provenance label attached by tools like ChatGPT, Gemini, Adobe Firefly, and Microsoft. RAIW reads the manifest and names the signing platform.
Which C2PA products can RAIW attribute precisely?
Besides issuer-based detection, RAIW recognizes signed product claims from Higgsfield AI, Topaz Labs Image API, and TikTok Ad Creative Toolbox, even when an upstream provider signs the credential.
How do I check for a SynthID watermark?
SynthID is an invisible watermark used by Google and OpenAI image tools. RAIW reads signed provenance: Google AI provenance and current OpenAI manifests with an explicit watermark action confirm SynthID; older OpenAI C2PA without that action does not. The pixel watermark itself is not locally decoded; confirm it in the Google Gemini app or at openai.com/verify.
What is the 'Made with AI' label and how do I detect it?
It is the IPTC digitalSourceType tag (and related C2PA data) that triggers AI disclosures on Instagram, Facebook, and X. RAIW detects this tag and tells you which platform conventions it matches.
Does the checker detect Doubao and other China AI images?
Yes. It reads the China AIGC label (TC260 standard) used by Doubao and other China-served generators, plus supported ByteDance-family app-export AIGC disclosures stored in EXIF metadata.
What does it mean if no signal is found?
It means RAIW did not find a supported provenance signal. The image can still be AI-generated - the signal may have been stripped on re-save, screenshot, or upload, or it may use a watermark the checker cannot read locally. Absence of a signal is not proof an image is human-made.
Is checking free? Do you store my image?
Checking is completely free and needs no account. Uploaded images and results currently have no automatic expiry. We retain them to run and improve the service until they are no longer needed or we receive a valid deletion request that identifies them, and we never sell them.
What image formats can I check?
PNG, JPG, WebP, AVIF, and HEIC files up to 150 MB.
I found a watermark - how do I remove it?
Use Quick clean to remove supported visible marks plus C2PA, Made with AI, and AIGC metadata. Use All-in-one to additionally regenerate pixels and target supported invisible markers such as SynthID for $1.49 per image. Quick clean is free with Standard output up to 12 MP. Keeping the original resolution above that size requires paid processing.
Invisible watermark removal starts at $1.49 per image, with no account required.