Generate image title, alt text and caption
The metadata tool looks at attachments and writes their title, alternative text and caption. It is the fastest way to deal with a media library where most images have no description at all.
Open it at Media → AI generate image ALT, caption and title.
You need permission to upload files and edit each selected image. Requests containing an image you cannot edit are rejected before generation starts. Duplicate image IDs in one request are processed once. Pomatio Platform keeps this tool available to Editors and Shop managers.
What the screen shows
A searchable list of your media library. Each image displays its existing title, alt text and caption, with a marker where a value is missing.
That display is the useful part: you can see at a glance which images have nothing, and select exactly those rather than reprocessing images that are already described.
The picker loads 24 images at a time. Selecting images updates a counter, and holding Shift while selecting another checkbox selects the whole visible range between the previous checkbox and the new one. You can select up to 48 images per batch — the tool refuses a larger selection.
Running it
Use the button that calculates credits before confirming. The estimate covers the whole selection; each image is a separate request and a separate charge.
After confirmation the tool processes one image at a time and updates the progress counter after each response. A short pause between requests helps avoid service rate limits. The Processing summary lists each image with the title, alt and caption it produced. When one image cannot be processed, the error is shown for that image and the remaining images continue.
What gets written
For each processed image:
- the title is written to the attachment title;
- the caption is written to the attachment caption;
- the alternative text is written to the attachment's alt text field.
An empty value returned for a field leaves that field as it was. Values that are produced overwrite what was there.
That last point matters: this tool overwrites existing metadata. If someone has already written careful alt text for an image, reprocessing it replaces their work. Select images that are missing values, not everything.
Batching sensibly
The 48-image limit is the maximum operation size, but smaller selections remain easier to review. A useful rhythm:
- Browse for images with no alt text.
- Select a reviewable group of similar images, up to 48.
- Estimate, confirm, run.
- Read every result in the summary.
- Correct the ones that are wrong before moving on.
Step 4 is the one people skip, and it is the one that determines whether this tool helped. A large group of descriptions takes real time to review.
Why grouping by kind helps
The tool describes what it sees, with no knowledge of where the image is used. Processing a batch of product photographs together makes the results consistent and easy to check. Mixing product shots, team photos, diagrams and decorative textures in one batch produces many unrelated descriptions to evaluate individually.
Where it works well and badly
Well: photographs with a clear subject — products, places, people, objects, food, interiors.
Badly: diagrams, charts, screenshots, infographics and any image whose meaning is its content rather than its appearance. A description of a chart's appearance is useless; what a chart needs is its conclusion. Write those by hand.
Not at all: the meaning an image carries in context. An image of a smiling person means something different on an About page and beside a testimonial. The tool cannot know which.
Alternative text needs review
Generated alt text describes what is in the picture. Good alt text conveys the purpose the image serves for the reader. These frequently differ, and decorative images should have empty alt text rather than a description.
Do not treat this tool as having solved accessibility. It has produced a first draft that is far better than nothing and not yet correct. See Review AI-generated alternative text.
Multilingual sites
Metadata is generated in the site's language. Translated posts share the same attachments, so a translated page carries alt text in the original language.
If that matters, describe images on each site or translate the values manually. Translating a post does not translate its images. See Preserve formatting, shortcodes and links.
Cost
One charge per image. A batch of 48 is 48 charges. A media library of 2,000 undescribed images is a substantial project — plan it in reviewable batches rather than as one campaign, and see What image operations cost.
A realistic plan for a large library
- Prioritise images that appear on pages that matter, not the whole library.
- Skip decorative images entirely; they should have empty alt text.
- Work in reviewable batches of no more than 48 images.
- Write diagram and chart descriptions by hand.
- Stop when the pages people actually visit are covered.