Image analysis

Let the AI look at your product photos so the copy it writes describes the real product.

By default, Product Pelican writes from your product data — title, type, tags, metafields. Image analysis adds your product photos to that, so the AI describes the product as it actually appears rather than inferring from text.

The difference is concrete. Without photos, a description says what the title implies. With photos, it can mention the ribbed cuff, the contrast stitching, the way the fabric drapes, and what is actually in the shot.

Turning it on

1. Go to AI Prompts.

2. Select the Description prompt.

3. Find the green Image analysis box.

4. Tick Use product images in AI generation.

5. Choose how many images to check per product — 1 to 5.

6. Save.

It is off until you turn it on. No photo is analysed and nothing counts against your allowance until you tick the box yourself.

What it affects

One switch, three outputs:

Output	What changes
Product descriptions	Colour, material, cut, pattern, styling and what's visible in the shot
Generated metafields	The metafields generated alongside the description are grounded in the same photos, so attributes stop being guesses
Image alt text	Alt text is written from the actual image rather than from the filename

Alt text always analyses exactly one image — the one it is describing. The image-count setting applies to descriptions and metafields.

Choosing how many images

Each image you analyse costs one unit of your monthly allowance, so the setting is a direct trade between quality and quantity.

Setting	Good for
1 image	Most catalogues. The primary shot usually carries the important detail
2–3 images	Products where a second angle matters — back detail, interior, texture, scale
4–5 images	Complex or high-value products where you want the fullest possible description

Images are taken in the order they appear on the product, so your primary image is always analysed first.

Your monthly allowance

Each analysed image counts once against a monthly limit set by your plan.

Plan	AI image analyses / month
Free	10
Standard	1,000
Pro	3,000
Max	6,000

The allowance resets on the 1st of each month.

Working out what you need

Multiply the products you plan to generate for by your image-count setting.

• 800 products at 1 image = 800 analyses — fits Standard.

• 800 products at 3 images = 2,400 analyses — needs Pro.

• Re-generating a product analyses its images again, so leave headroom for revisions.

When the allowance runs out

Generation carries on. It simply falls back to the text-only path — the same output you'd get with image analysis off — and the description panel shows a notice telling you the images were skipped.

Nothing fails and nothing is lost. Running out of allowance degrades quality for the rest of the month; it never blocks you from generating, and never produces an error you have to clear.

If you hit the limit regularly, either lower the image count or move up a plan — see Plans & billing.

Getting good results

• Your primary image should show the whole product. A cropped detail shot as image 1 gives the AI a partial view of what it's describing.

• Photos of packaging, size charts or lifestyle scenes are less useful as the first image — the AI describes what it sees, and a size chart is not the product.

• Re-tune your description prompt after turning it on. Prompt instructions written for text-only generation often produce over-long output once photos are added, because there is suddenly much more to say. See AI prompts.

• Use Show thinking on a few results to see what the AI actually picked up from the photo.

Related pages

• AI prompts — where the setting lives, and how to tune the description prompt.

• Plans & billing — allowances per plan.

• Product audit — where descriptions and alt text are generated.