How to Remove AI Metadata from a Single Image (Without Uploading It)

Step-by-step workflow for removing AI metadata from one image. Works on Firefly, Midjourney, DALL-E, Stable Diffusion. No upload, no account, free.

Sep 7, 2026

AI Metadata Ninja Team

The single-image case is the most common one. You generated an image, you want to publish it, and you want the AI trail gone before it leaves your machine. This guide walks through the exact workflow, including what to expect from the scan, what gets cleaned, and how to verify the result. The same tool also handles batches of images and the full verb-variant hub is here.

What "removing AI metadata" means in the single-file case

You have one JPEG (or PNG, or WebP). Inside the file is a payload of AI provenance — XMP parameters with the prompt, EXIF Software with the generator name, possibly a signed C2PA manifest. Removing the AI metadata means walking the file, finding those bytes, dropping them, and writing a new file with the rest of the bytes intact. The pixel data is untouched. The ICC color profile is preserved.

For one file, the whole process takes a few seconds. You drop the file in, you read the report, you click Clean, you download the result. The file never leaves your browser.

Step 1: drop the file

Open the homepage and drop the file into the drop zone. The drop zone is the central panel of the hero section — drag a file from Finder, or click to open a file picker. The file is opened in your browser via the File API. No request is made with the file contents; you can verify by opening the network tab in DevTools.

JPEG, PNG, and WebP are supported. HEIC is read-only — the scanner will report what is inside, but the cleaner cannot write HEIC. Convert to JPEG first if you need a cleaned HEIC.

Files up to 10 MB are accepted. The scanner loads the file into memory as a single ArrayBuffer and walks the bytes directly.

Step 2: read the report

Within a second, the scanner reports every AI-trailing field the file carries. The report is grouped by metadata standard:

C2PA — a signed Content Credential inside an APP11 marker. The report names the issuer (Adobe, Microsoft, etc.) and the actions recorded.
XMP — parameters (prompt, model, seed, CFG), dalle (DALL·E rewritten prompt), ideogram (Ideogram prompt), and similar blocks.
EXIF — the Software field, plus GPS coordinates if any.
IPTC — usually not AI-related, but checked for completeness.
PNG tEXt / iTXt chunks — the place Stable Diffusion and ComfyUI write their prompt when the export is PNG.

If the report says "no AI fields found", the file is already clean and there is nothing to do. You can still hit Clean — it will produce a byte-identical output, which is the cheapest possible verification.

Step 3: click Clean

The cleaner walks the file at the byte level. It finds the AI-trailing segments, removes them, and writes a new file. The pixel data is byte-identical to the source. The ICC color profile, if present, is preserved. GPS coordinates are removed by default (privacy).

The byte-level approach is what differentiates this from a re-save. Re-saving the file (Photoshop "Save for Web", macOS Preview "Export As", most online tools) re-encodes the JPEG, drifts the pixels, and can change color rendering. A byte-level walk only touches the metadata segments. The comparison is here if you want the full breakdown.

The result downloads as a new file. The original is untouched in your file system.

Step 4: re-scan to verify

The tool re-scans the cleaned file automatically and shows "0 AI fields remain" in the final report. You can also drag the cleaned file back into the drop zone to confirm yourself. The SHA-256 of the cleaned file is different from the source (because the metadata bytes are gone), but the pixel data is byte-identical.

This re-scan step is the proof. If a buyer downloads the cleaned file and runs their own metadata reader, they will see exactly what your scan showed: no AI trail.

What gets removed and what stays

| Field | Action | Why | | ------------------------------------- | ------------------ | ---------------------------------------------- | | XMP parameters, dalle, ideogram | Removed | Direct AI provenance — the prompt, model, seed | | EXIF Software if AI-originated | Removed | Generator name (e.g. Adobe Firefly 3.0) | | EXIF Software if camera-originated | Kept | Useful to the photographer | | EXIF GPS | Removed | Privacy default | | C2PA manifest | Removed by default | Strip on publish; keep only for auditability | | IPTC caption and copyright | Kept | Author credit | | ICC color profile | Kept | Print and color accuracy |

The cleaner is safe to run on any file. If the input is already clean, the output is byte-identical. If the input has AI metadata, the output has only the AI metadata removed.

When to use the single-file flow vs the batch flow

Single-file is the right choice when you are working on one image, or when the images you are working on need different decisions (some keep the C2PA credential, some strip it). Drop them in one at a time and decide per file.

For ten or more files with the same cleaning policy, the batch workflow is faster — drop them all into the same drop zone, and the scanner processes them in sequence. The free tier covers 10 files a day anonymous and 20 once you sign up. For higher volume, the paid credit packs give you 100, 500, or 1500 cleanings inside a month.