What would it cost to tag your whole library with AI?

For each image we add the input-token cost (the image plus the prompt) and the output-token cost (the tags or caption returned), using the per-million rates you set. That per-image figure multiplies by your library size, the images analysed per item, and the number of passes. All the math runs in your browser.

A cheap first pass captions the media and pulls broad tags. A second pass does the precise work: matching the frame to your product catalogue and scoring quality so weak UGC never reaches a shopper. You can switch to a single pass if you only need captions and tags.

No. The input and output rates are editable fields, defaulted to a sensible Haiku starting point. Model rates move, so overwrite them with the published rate on the day for an accurate estimate. Nothing here is a frozen pricing claim.

Treat it as a ceiling for back-of-envelope planning. Real cost depends on image resolution (which changes the token count per image) and the rate you pay. Idukki batches requests and caches results across a library, which pulls the real number below a naive per-image estimate.