EN

Clean AI images

Strips C2PA, prompts and EXIF without changing a single pixel.

FreeNo account

Tool

Next stepAI Image Detector

How It Works

  1. 01

    Drop images

    Pick single images or a whole folder. PNG, JPG and WebP are recognised.

  2. 02

    Check the origin

    Each image shows whether it declares itself as AI-generated, and by which generator.

  3. 03

    Clean and download

    One click removes the metadata. Download one by one or as a ZIP.

Privacy

Everything runs in your browser. No uploads, no storage.

AI images often carry more than you think: the full prompt, a ComfyUI workflow with the file names of your models, or a C2PA manifest with a timestamp. This tool shows you that data and removes it losslessly, whole folders included. An invisible watermark inside the pixels stays where it is.

01: How to Use

How do you use this tool?

  1. Drag images into the drop area, or pick a whole folder.
  2. Check which provenance data each image carries.
  3. Click “Remove metadata” and download the images one by one or as a ZIP.

What does an AI image carry besides the picture?

More than most people expect when they share one. ChatGPT writes a C2PA manifest into its exported images; an export from 29 September 2026 carried about 21 KB of it, with program name, timestamp and certificate chain. Stable Diffusion interfaces put the prompt, negative prompt, seed and sampler into a PNG text chunk. ComfyUI saves the complete workflow as JSON, including the file names of your models. Post such an image in a forum or send it to a client and all of that goes along.

What exactly does the tool remove?

Every block that describes origin or capture: the C2PA manifest, text chunks with prompts and workflows, XMP, EXIF, IPTC and JPEG comments. The manifest sits in a different place in each format, in the caBX chunk of a PNG, in APP11 segments of a JPEG and in the C2PA chunk of a WebP. The tool handles all three.

Two things stay: the ICC color profile and, for JPEG, the orientation tag. Without them colors would shift and portrait photos would turn sideways.

After cleaning, the tool reads the new file again with the same check the AI image detector uses. It only reports “cleaned” once no manifest turns up. If something is still there, the list says so in red.

Why does the image quality stay the same?

The tool never touches the image data. It cuts the metadata blocks out of the file and puts the rest back together unchanged. Nothing gets re-encoded, so you get no JPEG artifacts and no color shift. The file shrinks by exactly the bytes removed, a little over 21 KB for a ChatGPT manifest.

Can people still tell it is an AI image?

Often they can. Many services also put an invisible watermark straight into the pixels, Google for example uses SynthID. That watermark lives in no metadata block, cleaning leaves it alone, and this tool does not try to remove it.

Cleaning does not change any labelling duty either. Article 50 of the EU AI Act applies from 2 August 2026: whoever publishes a deepfake must disclose that it is artificial, whatever the file contains.

When is cleaning worth it?

Mostly for prompts and workflows you do not want to hand out. A client gets the finished picture without your recipe, and a forum does not learn which models sit on your machine. Timestamps are another reason, since they show when you worked on what. File size is the last one: 21 KB of manifest is a noticeable share of a small preview image.

With a whole folder, the summary at the top shows how many images carry AI provenance data at all. Images without metadata stay untouched.

Which tools go with it?

The AI image detector shows the provenance data of a single image in full, with time, evidence and prompt. Remove metadata also handles PDF files and lets you keep individual fields. The EXIF viewer shows all camera data of a photo as a searchable tree.