AI Metadata Tag Pack
Write the IPTC digital source type and keywords into up to 250 JPEG, PNG and WebP images in one batch. Add the images as files, a pasted list of links, a CSV file or a Merchant Center product feed. It runs in your browser without re-encoding a pixel, and every file is read back after it is written. You get ZIP files of the tagged images and a manifest CSV file that records, for each image, what was written or why it was not, with the SHA-256 of the file before and after.
Published . Last reviewed .
$19 for up to 250 image files within 30 days of paying. One payment, nothing renews. Your images are never uploaded to GotAlt. How to check that for yourself.
GotAlt reads what a file and a page declare. It cannot tell whether an image was made with AI.
Your pack
Tag your images
1. What should the files declare?
Write only what is true of every image in this batch. This page does not know how your images were made, and it does not suggest a value. A CSV file can give one row its own values.
Choose the value that is true of the images.
Added to the XMP keyword list (dc:subject). Keywords a file already has are kept.
Amazon's reported keyword for images of photorealistic people generated entirely by AI. Not for real people edited with AI, and not a visible label. What is known about Amazon's rule.
Written as Iptc4xmpExt:AISystemUsed. The name of the tool or model, as you want it recorded.
Files that carry Content Credentials (C2PA)
A file with a C2PA manifest already declares how it was made, in signed form. Writing XMP into it changes bytes that the manifest's hash covers, so the signature becomes invalid. By default the pack leaves such files unchanged and lists them as refused. The manifest itself is never removed, even if you override this.
2. Add your images
Drop image files here
or press Enter to choose files from your computer
JPEG, PNG or WebP, 50 MB each. Written in your browser, never uploaded. A file whose name matches a link or a CSV row that could not load takes that row's place.
Your browser fetches each link straight from the host it names, not through GotAlt. Many hosts do not allow that. Download and drop any file that does not load.
A header row with a column named image_url (a link, or the name of a file you will add). Optional columns source_type, keywords and ai_system give that row its own values. The columns are listed further down this page.
A CSV, TSV or XML feed with an image_link column. You choose which products' images to tag.
3. Check the list
| Image | From | Status | Detail |
|---|
4. Tag the images
Your files
Order the pack
What ends, and what is deleted, when the 30 days end
- What ends. 30 days after you pay, the pack stops tagging images, and this page shows the order form again. Images already tagged and downloaded are not affected.
- What is deleted. Nothing of yours is held by GotAlt, because your images, their links and their file names never reach it. This page removes the order number and the count of tagged images that it keeps in your browser's local storage the next time it sees that the pack has ended.
- What stays. The order record GotAlt keeps with its billing records: what you bought, when, what you paid and the two boxes you ticked. The ZIP files and the manifest you downloaded stay on your computer, and they are yours.
Already bought a pack?
What you get
The free AI image metadata writer takes one batch of files at a time and stays free. The AI Metadata Tag Pack is for catalogs: $19 for one payment buys up to 250 image files within 30 days of paying, tagged in this page.
- Batch input. Drop files, paste a list of image links, choose a CSV file with one image per row, or choose a Merchant Center product feed and pick the products whose images to tag.
- The same writer as the free tool. Each image gets the IPTC digital source type you choose and the keywords you enter, in its XMP block. No pixel is decoded or re-encoded, and nothing is ever removed.
- A verify pass. After a ZIP is built, the page opens it again and reads every file back. A file whose source type, keywords or checksum does not read back as written is refused, left out of the ZIP and not counted.
- ZIP files. The tagged images are packed into ZIP files of about 100 MB each at most, so a large batch is never one huge file.
- A manifest. One CSV row per image, kept as your record. A file name, keyword or link that starts with =, +, - or @ is saved as text, so a spreadsheet does not run it as a formula.
What counts as one image
An image counts once it has been written and read back from its ZIP. These do not count: an image the page refused (a file with Content Credentials, a format other than JPEG, PNG or WebP, a link that could not be fetched), a file that already carried what you asked for, and a file the verify pass refused. Images are tagged one after another in the order of the list, and the page stops at the number the pack has left, so you can run the rest of a list later. The count is kept in this browser and applied by this page, so it carries over when you reload the page and is not shared with another browser or device.
Adding images
Files
Drop or choose JPEG, PNG and WebP files of up to 50 MB each. GIF, AVIF, HEIC, TIFF and SVG files are refused, and no image is converted to another format. A file added twice is added once.
Links, and why one may not load
Paste one https link per line. Your browser fetches each link from the host it names, and a browser may read an image from another site only when that site allows it (CORS). Some hosts allow it and many do not. A link that does not load is listed as Not fetched, with the reason, and you add the downloaded file with the same file name: it takes the row's place. This page never promises that a link will load.
A link gives you the file the host serves, and some hosts serve a resized copy that has lost its metadata. Whenever you have the original file you upload to your store, tag that file. The AI image metadata checker compares what a link delivers.
A CSV file
The first row names the columns. A file with a single column of links and no header row is read as a list of links. A row with a value in a column of its own uses that value for that image; an empty cell uses the batch choice above.
| Column | Needed | What it holds |
|---|---|---|
| image_url | Yes | An https link, or the name of a file you will add. image_link, url, src and file_name work too. |
| source_type | No | An IPTC term such as trainedAlgorithmicMedia. It replaces the batch choice for this row. A value that is not a current IPTC term refuses the row. |
| keywords | No | Keywords separated by commas or semicolons. They replace the batch keywords for this row. |
| ai_system | No | The AI system used. It replaces the batch value for this row. |
A Merchant Center feed
Choose a feed file in CSV, TSV or XML form. The page reads its image_link column, and optionally additional_image_link and lifestyle_image_link, and you choose all products, the products whose IDs you list, or the products where a column has a value. Google's help page on preserving metadata for AI-generated images says that all AI-generated images must contain the IPTC DigitalSourceType trainedAlgorithmicMedia tag and that embedded metadata should not be removed (read October 22, 2026). The Merchant Center AI feed checker reads what a feed and its image files declare.
The manifest
The manifest lists every image in the list in the order it was added, whether it was written or not. A result is one of Written, Unchanged (the file already carried what you asked for, so no new file was made), Refused (with the reason), Not fetched (a link the browser could not load) or Not processed (the pack ran out of images, the 30 days ended, or you stopped the run). It is a record of what the page found and wrote. It is not a statement about how any image was made.
| Column | What it holds |
|---|---|
| File name | The file's name, or the last part of the link. |
| Result | Written, Unchanged, Refused, Not fetched or Not processed. |
| Reason or notes | Why a file was refused or not fetched, or what the writer noted for a written file, such as a WebP header change. |
| Source type written | The IPTC term added to the file. |
| Keywords written | The keywords added, separated by semicolons. Keywords the file already had are not listed. |
| SHA-256 before | The checksum of the file as the page received it. |
| SHA-256 after | The checksum of the tagged file, read back from its ZIP. The same as before for a file left unchanged. |
| Output file | The name of the tagged file inside the ZIP. |
| ZIP part | Which ZIP file holds it. |
| Read back from the ZIP | yes when the verify pass found the file in the ZIP and read back what was asked. |
| Pixel data unchanged | yes when the writer proved that only the XMP block differs from the original. |
| Source type requested | The value asked for, from the batch or the CSV row. |
| Keywords requested | The keywords asked for. |
| Where it came from | file, link, or file added for a link. |
| Link | The link, for an image given as one. |
| Reference | The CSV row or the feed product the image came from. |
| Bytes before | The size of the file as received. |
| Bytes after | The size of the tagged file. |
| Processed (UTC) | When the page handled the image, in UTC. |
After you upload the tagged files
This pack writes AI metadata into your image files. Once they are on your site, the Site Image Check reads the images on your live pages, including what each image file declares about AI.
What the pack does not do
- It writes what you choose. It does not know how an image was made and cannot tell. A source type that is not true of an image is a false statement in your file.
- It never removes anything. There is no remove or clean-up option. Existing keywords, the creator tool and every other field stay, and a declaration of generative AI is never replaced by a value that does not say so. The EU Code of Practice (Measure 1.2, June 10, 2026) says its signatories will make best efforts to preserve metadata markings and will not promote tools whose purpose is to circumvent machine-readable markings.
- It does not re-encode. It copies every byte of the image data unchanged and adds or edits only the XMP block. Before a file goes into a ZIP the writer compares it with the original, part by part, and refuses a file where anything else differs.
- It leaves signed files alone by default. A file with a C2PA manifest is refused unless you tick the box for Content Credentials, because writing XMP into it makes its signature invalid.
- It cannot make a platform read the tag. A marker in a file you upload is not a statement that a platform reads it, and many hosts drop markers when they resize an image. See which platform reads which field.
- It is not legal advice and does not make anything compliant. Metadata is one way some platforms ask for AI content to be marked. It is not a visible label, and nothing here says it satisfies any law. See the AI disclosure laws finder for rules that may apply to you.
Refunds and what we keep
Refunds
If the pack could not write to your images, or you find that most of its results are wrong, email hello@gotalt.com from the address you paid with within 7 days of paying, and we refund the payment in full.
What we keep
Paying creates a GotAlt account with the email address you give at checkout, unless that address already has one, and the pack goes on that account. We keep the order: what you bought, when, what you paid and the two boxes you ticked, with our billing records. Your images, their links and their file names never reach us. Our privacy policy has the details, and our terms of service cover the pack in full.
So that the pack stays unlocked in this browser, the page keeps your order number and the count of images tagged in your browser's local storage on this site. If that is cleared, open the link in your confirmation email, or sign in and choose Unlock my pack in this browser.
Questions
Are my images uploaded?
No. Files are read and written in this browser tab by script files you can read: tag-pack.js, xmp-write.js, provenance.js and zip-store.js. The one script that talks to GotAlt, ai-tag-pack-ui.js, handles the order and never reads a file. If you paste image links, your browser fetches each one from the host the link names, and that host sees the request. To check for yourself, open your browser's developer tools, switch to the Network tab and tag a file: apart from this page's own files, nothing is requested.
What happens when the 30 days end?
The pack stops tagging images and this page shows the order form again. The order number and the count that the page kept in your browser are removed. The ZIP and manifest files you downloaded stay on your computer. GotAlt keeps the order record with its billing records and nothing else, because it never held your images.
Does the free writer still work?
Yes. The free writer tags images one batch at a time with no limit on how many you tag over a day. Its ZIP holds up to 10 files per batch, and every file can still be downloaded on its own. It does not make a manifest. This pack adds batch input, the verify pass, ZIP files of the whole batch and the manifest.
Why was an image refused?
Each row in the list and in the manifest says why in plain words. The usual reasons are a file with Content Credentials, a file that is not a JPEG, PNG or WebP, a file that is cut off, XMP in a shape the writer will not rewrite, a CSV row with a source type that is not a current IPTC term, and a file over 50 MB.
Does a tagged image prove how it was made?
No. A tag is what the file says about itself, written by you. A file with no tag proves nothing about how an image was made, and GotAlt cannot tell whether an image was made with AI.
Related
- AI image metadata: what C2PA, IPTC and XMP keywords are and how they travel.
- AI image metadata writer: the free tool for one batch at a time.
- AI image metadata checker: read what a file declares and what survives your store.
- Merchant Center AI feed checker: check a product feed's AI attributes and sampled image tags.
- Which AI disclosure rules apply? What may apply, with sources and dates.
Metadata is one check; the page around the image is another
The free scan runs 16 common checks on up to three pages of your site in under a minute: missing alt text, form labels, headings, page language, link text and more.
Scan my site for freeThis is not legal advice. You remain responsible for the disclosures you publish.