AI Image Sanitizer

Drop an AI-generated PNG, JPEG or WebP and remove the embedded prompt, seed and workflow data, EXIF, XMP and C2PA Content Credentials. Optionally add invisible noise, trim the edges or rescale slightly to perturb pixel-level patterns, compare before and after, and download the cleaned image. Everything runs in your browser.

PrivateFreeWorks offline

Reviewed by Vincenzo Manto, Founder & Lead Developer · Last updated October 1, 2026

How it works

  1. 1

    Drop an AI-generated image (PNG, JPEG or WebP).

  2. 2

    See which metadata and Content Credentials the original carries.

  3. 3

    Pick the output format and tune noise, edge crop and rescale.

  4. 4

    Compare the result, check what is left in it, and download.

Share images without the prompt attached

AI generators embed the full prompt, seed and sometimes the entire node graph in the file. This tool rebuilds the image from its pixels so none of that travels with it, and shows you what was found.

Where AI images keep their hidden data

Stable Diffusion front ends write the prompt, negative prompt, seed, sampler and model hash into PNG text chunks; ComfyUI stores the whole node graph as JSON; Midjourney and DALL-E outputs may carry XMP fields and C2PA Content Credentials that name the generator. JPEG and WebP files can hold the same data in EXIF and XMP segments. The reader panel lists every block found before you clean anything.

What re-encoding does and does not change

Redrawing the image onto a canvas and exporting it again discards every metadata container, because only pixel values are copied. It does not alter what the picture shows. Pixel-level options such as noise, a few pixels of edge crop or a 1-2% rescale add small differences that can break naive hash matching, but they are not a guarantee against statistical watermarks.

When this is the right tool

Use it before posting a render publicly, attaching it to a client brief, or uploading it to a stock site that asks for no embedded prompt data. If you only want to inspect a file without altering it, the AI Image Metadata Reader shows the same fields read-only.

Frequently asked questions

Is my image uploaded anywhere?

No. Decoding, cleaning and re-encoding all happen in your browser, so the image never leaves your device.

What exactly gets removed?

The image is decoded to raw pixels and re-encoded, which drops EXIF, XMP, PNG text chunks (prompts, seeds, ComfyUI workflows) and C2PA Content Credentials. The tool re-reads the output to show what is left.

Does it defeat invisible watermarks like SynthID?

Not reliably. Noise, cropping and rescaling perturb the pixels, but robust watermarks are designed to survive such changes, so treat the result as best-effort.

Which format should I choose?

PNG is lossless and keeps the noise exactly; JPEG and WebP are smaller and their re-compression also blurs fine detail.

Related tools

AI Music Sanitizer

Strip metadata and ultrasonic AI signatures from audio tracks.

AI Token Counter & API Cost Estimator

Estimate how many tokens your text uses for GPT, Claude and Llama, and what it will cost.

JSONL Validator & Converter (CSV, JSON ↔ JSONL)

Validate JSON Lines files for AI fine-tuning and convert between JSONL, CSV and JSON arrays.