AI tagging
Tag the products in your UGC, automatically.
Gemini vision matches each post image to products in your synced catalogue. By default, matches at 0.75 confidence or above are tagged automatically and the 0.5 to 0.75 band waits for a person. You can adjust both thresholds.
No code. Free plan, no card.
The moment
A thousand customer photos arrive and not one of them is linked to a product. Tagging them by hand is a week nobody has, so the gallery stays pretty and unshoppable.
What Idukki does
Vision matching proposes the product for each post; confident matches tag themselves and uncertain ones wait for a person, so nothing wrong goes live on a guess.
01
Shortlist, then match
Large catalogues are narrowed first: the post’s caption and labels are embedded and the 60 closest products are sent with the image. The model can only answer with products from that list, so a match is a real SKU, not a guess.
Shoppable UGC
A pin becomes an order
WROGN Men Silver-Toned Watch
4.8 · 213 reviews
$24.76
Orders ledger
#4821$24.76attributed to@darcy.s
02
Confidence bands, not a coin flip
By default, matches at 0.75 confidence or above are tagged automatically, matches from 0.5 to 0.75 wait in the Recommendations queue with the product attached, and anything below 0.5 is dropped. Tagging too much or too little? Move either threshold on the Recommendations page, within 0.3 to 0.98, with the review floor always below the auto-tag line.
FRAME · 0:08UGC reel · @priya03
One catalogue for every source
Posts from Instagram, TikTok, YouTube, reviews and uploads are all matched against the same synced Shopify, WooCommerce or BigCommerce catalogue, so a tag resolves to a real product page and a real price.
Sources
Nine sources, one feed, one script
- TikTok
- YouTube
- X
- Threads
- Google Reviews
Gallery · homepage
<script src="…/idukki.js"> · 37 KB- TikTok
- YouTube
- X
- Threads
04
Hotspots timed to the video
On video, the tagger reads the thumbnail. In the hotspot editor you can give a product tag a timestamp, so the pin and product card appear at the second the product is on screen.
Widget builder
Pick sources and a layout; the preview is the real renderer
Sources
- TikTok
- YouTube
- Hashtag
- Reviews
Layout
Products
AI tagging on
Add to cart inline
Live preview
How it works, from switched on to paying for itself.
- Step 1
Ingest the asset
From Instagram, TikTok, YouTube, the DAM, a hashtag campaign or a creator upload. Photos use the image; videos use the thumbnail.
- Step 2
Shortlist
Over 150 products, the caption and labels are embedded and the 60 closest products are picked. Smaller catalogues send every product.
- Step 3
Match
One Gemini vision call returns the catalogue products it can see, each with a confidence score.
- Step 4
Route
Above your auto-tag threshold (0.75 by default) is tagged. Between that and your review floor (0.5 by default) waits in Recommendations. Below the floor is dropped.
“Being able to see how the clothes move and look on different women makes a huge difference.”
Three things to know about how the tagger decides.
The middle band waits in the Recommendations queue with the candidate product already attached, so a person confirms in one click.
- 01
Shortlist first, then one vision call
For catalogues over 150 products, the post’s caption and labels are embedded and the 60 closest products go into the prompt. Then a single Gemini vision call picks the ones it can actually see.
- 02
Three confidence bands, set by you
By default, matches at 75% confidence or above are tagged automatically, matches from 50% to 75% wait in the Recommendations queue for a person to accept, and anything lower is dropped. You can move both lines on the Recommendations page: the review floor can go down to 30%, the auto-tag line up to 98%, and the floor always stays below the auto-tag line. Reset puts the defaults back.
- 03
It only picks from your catalogue
The model can only return products from your own synced catalogue. Any id it invents is thrown away, and matches are capped per image.
The questions to ask before you switch it on.
Which catalogues can it map to?
Any synced catalogue: Shopify, WooCommerce, BigCommerce, Wix, or a CSV/API feed for custom stores.
What happens to lower-confidence matches?
Matches from 0.5 to 0.75 go to the Recommendations queue with the candidate product attached, for a person to accept or reject. Matches below 0.5 are dropped. Those are the defaults: you can set your own review floor and auto-tag threshold on the Recommendations page, anywhere from 0.3 to 0.98, as long as the floor stays below the auto-tag line, and reset them at any time.
Does it work on video?
Yes, from the video’s thumbnail. If you want the product to appear at a particular second, set a timestamp on the tag in the hotspot editor.
Which model does it use?
Google Gemini vision (currently Gemini 3.6 Flash), with Gemini embeddings for the catalogue shortlist.
Is AI tagging on every plan?
Product tagging on image and video, multiple product tagging and hotspot tagging are listed for every band, including Free, in the comparison table at /pricing. The AI Tagging API is also available standalone on the API platform.