AI image detector: is this image AI? Free check

How the check works

1
Content Credentials and IPTC
We read the file's C2PA manifest and XMP metadata for the IPTC digital source type. “Trained algorithmic media” means a generator made it; “composite with trained algorithmic media” means AI edited a photo.
2
Generator fingerprints
Stable Diffusion, ComfyUI, InvokeAI and NovelAI write their settings into PNG files, and many tools name themselves in the software field. We read every text chunk and EXIF field.
3
Camera data
Real cameras record make, model, lens and exposure settings. We read them all and weigh them with everything else.
4
Close visual inspection
A vision model inspects the image for the tells generators leave: hands, teeth, text, reflections, repeated backgrounds and over-smooth skin.

What each AI image generator leaves in its files

Most of the big generators now label their images. The label survives as long as nobody screenshots, crops or re-saves the file.

Generator	What it writes into the image	What Gireta reads
ChatGPT and GPT Image (OpenAI)	A signed C2PA manifest naming OpenAI	The manifest and its source type
Gemini and Nano Banana (Google)	A SynthID watermark in the pixels, plus C2PA and the IPTC source type	C2PA and IPTC; SynthID needs Google's own detector
Adobe Firefly	C2PA Content Credentials, since launch	The manifest and its source type
Microsoft Designer and Bing Image Creator	C2PA Content Credentials	The manifest
Midjourney	No C2PA; committed to the IPTC source type in 2023	IPTC fields and prompt text when present, then the visual check
Stable Diffusion (Automatic1111, Forge, Fooocus)	A PNG text chunk called parameters: prompt, seed, sampler, steps	The whole block, word for word
ComfyUI	The prompt and workflow node graph as JSON in the PNG	The workflow
Flux, Grok and most others	Usually nothing	The visual check and camera data

Screenshots, social apps and most image editors strip all of this. A missing label does not mean a real photo.

What the verdict means

Verdict	What decided it
AI-generated	The file's own provenance or generator settings say AI made it. This is as certain as it gets.
AI-edited	Provenance says a photo or image was changed with generative AI.
Camera original	Content Credentials record a camera capture.
Likely AI	No provenance, but the visual check and other clues lean towards AI.
Likely a real photo	No provenance, but camera data and the visual check lean towards a photo.
Mixed signals	The evidence points both ways. Check the original file or run a reverse image search.

Why Gireta catches what single-model detectors miss

Most detectors run one classifier over the pixels and hand back a score. Gireta starts where the proof is: the signatures AI tools embed in their own files. When ChatGPT, Gemini, Firefly, Sora or Stable Diffusion have labelled an image, Gireta reads that label, confirms the verdict and names the tool.

When the labels have been stripped, four more layers run: generator fingerprints in the file, camera data, a close visual inspection of hands, text, light and texture, and forensic views of the pixels. The verdict comes with the evidence behind it, so you can show your work.

And it all runs in your browser, so you can check sensitive images without uploading them anywhere.

SynthID and this detector

SynthID is Google's invisible watermark. Google opened its SynthID Detector to everyone in English on 7 October 2026. It is free with a Google, Apple or OpenAI sign-in and reads watermarks from Google, OpenAI, NVIDIA and Kakao models. Google has not published accuracy figures.

Gireta reads the open standards (C2PA and IPTC) and the generator settings in the file; SynthID Detector reads Google's watermark in the pixels. Run both for complete coverage of Google, OpenAI and open-source images.

More ways to check

Images

How to tell if an image is AI generatedTen ways to tell if an image is AI-generated, from the strongest evidence (Content Credentials, metadata, SynthID) to the visual tells in hands, text and light.Content Credentials (C2PA) checkerCheck an image's Content Credentials (C2PA) and IPTC source type online, free. See if it was made or edited with AI, and which tool made it.AI image metadata checkerSee an image's hidden metadata: the Stable Diffusion or ComfyUI prompt, EXIF camera data, XMP and C2PA. Free, in your browser, with error level analysis.ChatGPT image detectorCheck if an image was made with ChatGPT, GPT Image or DALL·E. Reads OpenAI's Content Credentials and inspects the image, free and private.Gemini and Nano Banana image detectorCheck if an image was made with Google Gemini, Nano Banana or Imagen. Reads C2PA and IPTC labels in your browser and explains SynthID.Midjourney detectorCheck if an image was made with Midjourney. Reads IPTC labels and prompt text, then inspects the image for Midjourney's look. Free and private.Stable Diffusion detectorCheck if an image was made with Stable Diffusion or SDXL and read its prompt, seed, sampler and model from the file. Works with Automatic1111, Forge, Fooocus and ComfyUI.Flux image detectorCheck if an image was made with Flux. Reads ComfyUI and Forge workflows from the file, names the model, and inspects the image. Free and private.Grok image detectorCheck if an image was made with Grok on X. Reads any embedded labels and inspects the image for the tells of AI generation. Free and private.

Video

AI video detectorCheck if a video is AI-generated, free. Reads Content Credentials and encoder tags from Sora, Veo, Kling and others, then compares frames for the tells.Sora detectorCheck if a video was made with OpenAI's Sora. Reads Sora's Content Credentials in your browser and compares frames for the tells. Free and private.Veo detectorCheck if a video was made with Google Veo. Reads embedded labels in your browser and compares frames for the tells of AI video. Free and private.Kling AI detectorCheck if a video was made with Kling AI. Reads encoder tags and compares frames for the tells of AI video, even when the Kling logo is cropped. Free.

Make AI images that stand out

Gireta puts Nano Banana, GPT Image, Seedream and every top image model in one studio, with one balance.

Start creating

Questions

Is the Gireta AI image detector free?

Yes. There is no sign-up and no limit on the file checks. The visual check is rate limited per visitor to keep it free.

Do you keep my image?

No. The file is read in your browser and never uploaded. For the visual check, a small preview (at most 1024 pixels wide) is sent, inspected and discarded.

Can it tell me which AI made the image?

Yes. It reads the signature AI tools leave in their files and names ChatGPT, Gemini, Adobe Firefly, Midjourney, Stable Diffusion, ComfyUI, Sora, Veo and more.

How accurate is the Gireta AI image detector?

When an image carries its generator's signature, the result is confirmed: the file itself says how it was made, and we name the tool. Without one, every independent signal is weighed together and the verdict comes with a confidence level, from moderate to high.

How do I check if an image is AI-generated?

Drop it into the Gireta AI image detector above. It checks the file's Content Credentials and IPTC source type first, which settle the question when present, then generator settings and camera data, then the pixels. For a second opinion, run Google's SynthID Detector and a reverse image search.

Can it detect ChatGPT, Gemini and Midjourney images?

ChatGPT and Gemini images carry Content Credentials, which it reads directly. Midjourney images carry no C2PA, so the result rests on IPTC fields when present and on the visual check. Any of these lose their labels once screenshotted.

Does a screenshot fool the detector?

A screenshot strips the labels in the file, but not the tells in the picture. Gireta's visual inspection and forensic views still run on every image.

Is an image with camera EXIF data real?

It leans that way, but EXIF is easy to copy onto any file. Treat it as supporting evidence, and look for Content Credentials from the camera for proof.

Sources

Last updated October 19, 2026.

AI image detectorIs this image AI?