AIGCTrace is a free AI image provenance tool. Inspect metadata, watermark clues, screenshot features and statistical signals. The 0–100 risk score is not a probability of AI generation.
Four core signal categories plus screenshot and statistical checks produce six report entries.
Finds C2PA markers, AIGC fields and tool names such as ChatGPT, Gemini, Doubao and Qwen. It does not validate cryptographic signatures or certificate trust chains.
Checks metadata, filenames and readable watermark-related markers. A SynthID text match is not pixel-watermark verification; this tool does not decode proprietary watermarks.
Flags periodic patterns in LSB bit planes without decoding watermarks. Compression, texture and screenshots can affect results.
Uses edge density and contrast to flag text-like overlays. It does not perform OCR or establish AI origin from text regions alone.
No sign-up needed. Upload directly, results in seconds.
Drag & drop or click to upload. PNG / JPEG / WebP / GIF / TIFF, up to 15 MB.
Checks provenance markers, watermark clues, screenshot features and statistical signals. Unavailable checks are explained in the report.
A 0–100 composite score with per-layer confidence and evidence. Export a shareable report image with one click.
Anyone who needs to verify the authenticity or origin of an image.
Check whether editorial images carry AI-generated markers to reduce misinformation risks.
Check whether readable provenance markers remain and compare with the original file.
Pre-screen uploaded images and quickly flag high-risk AIGC content to improve moderation efficiency.
Study C2PA adoption and cross-platform watermark retention and detectability.
Read the evidence before the score. Provenance markers, pixel features and statistical anomalies answer different questions.
When C2PA or tool names appear, check original files and source records. This tool does not validate signatures, and TC260 field recognition is not a conformity certification.
Screenshots, conversion and platform compression can remove clues. The online app skips pixel analysis above a 3000-pixel long edge and for WebP, GIF and TIFF.
Scores organize evidence rather than measure accuracy. Screenshot features and statistical signals are also heuristics that require source context.
C2PA specifications · Google DeepMind SynthID · AIGCTrace source code
Content reviewed: 2026-10-08 against project source; not a deployment date or independent accuracy benchmark.
Practical steps for checking evidence, from obtaining the original to interpreting a report.
AIGC means AI-generated content. AIGCTrace reports six checks: C2PA markers, readable metadata, visible watermark clues, blind watermark clues, screenshot features and statistical signals. Matching a tool name is not validated detection coverage for every version or image from that tool.
C2PA is a technical standard for content provenance and editing history. AIGCTrace searches for related markers but does not validate cryptographic signatures, certificate trust chains or the complete editing history. A marker alone proves neither credential validity nor AI origin.
Visible watermarks are text or logos in the image; invisible watermarks can be embedded in pixels. Blind detection does not require the original unwatermarked image. Metadata markers and pixel watermarks differ. AIGCTrace checks readable markers and pixel heuristics, without general-purpose decoding of proprietary watermarks such as SynthID.
Risk scores and confidence values are heuristic indicators, not calibrated probabilities or accuracy measurements. Metadata can be removed or forged; compression and texture affect pixel checks. Review each check and its limitations alongside original files and source context. No detected signal does not prove a camera-captured image.
There is currently no method that can determine with 100% certainty whether an image was AI-generated. AIGCTrace provides origin-signal detection — analyzing C2PA metadata, digital watermarks, and other embedded signals to determine whether an image carries markers from known AI tools. The absence of AIGC markers does not prove authenticity; strong signals indicate a high likelihood of AI origin. We recommend combining results with source context and other verification tools.
Original uploads and plaintext filenames are not stored. Images are processed in server memory; usage records may include file hashes, results and coarse request context. IP and User-Agent values are salted hashes by default; self-hosted configurations can enable raw-value logging.
The online app accepts PNG, JPEG, WebP, GIF and TIFF up to 15 MB. Pixel analysis requires decodable PNG/JPEG and is skipped above a 3000-pixel long edge; other formats can still be checked for readable metadata. Video and audio are outside the current online app scope.
Google’s official SynthID Detector (synthid.com) opened to everyone in October 2026. It checks images, video and audio made with AI from Google and partners such as OpenAI, NVIDIA and Kakao; Gemini also offers verification. AIGCTrace does not decode SynthID pixel watermarks, but checks the same image for C2PA, generator names and China AIGC labels for free, covering sources SynthID cannot. A SynthID text match is not watermark verification; no match does not prove camera capture.
Upload yours and see what it reveals. Free, no sign-up, results in seconds.
