Visual Product Discovery lets your buyers find products with an image. Your Bagisto store reads the picture with AI-Powered Search and shows matching products.

Online stores began with categories. A buyer clicked Men, then Jeans, then scrolled until something looked right.
Then came the search box. It works well, as long as the buyer knows the right words.
That is the catch. A buyer sees a jacket on a friend and wants one like it. To type a search, they first have to answer questions they cannot:
Every wrong guess is an empty results page. An empty results page is where a sale ends.
The solution is visual product discovery in Bagisto.
Bagisto already supports visual product discovery on the storefront. A camera icon sits at the end of the search bar, on desktop and on mobile.
![]()
Generative AI Image Search keeps the same camera icon and changes the engine. Instead of a general label, generative AI looks at the image.
It writes down what it sees, the way a shop assistant would: the product type, color, material, style and fit.
If a brand is visible on the product, the AI adds it to the keywords too.
In Bagisto, it is part of Magic AI, the built-in module behind every generative AI tool in the store. The admin panel calls it AI Image Search.
It works with eight AI providers, including Anthropic, OpenAI and Gemini, or with open models on your own server through Ollama.
Both engines power visual product discovery and answer the same question, “what is in this image?”, at a different depth. You choose which one.

Turn AI Image Search on, and the store uses it. Leave it off, and the store keeps using TensorFlow, so the camera icon always works.
The same fallback kicks in if the AI provider is ever unreachable, so buyers never hit a dead end.
Selling through several channels? Each channel sets its own switch and model, so one can use Gemini while another stays on TensorFlow.
For the buyer, an image search takes three steps:
Here is a real search: an outdoor image of a man in jeans, a denim jacket and suede sneakers. The AI returned eight keywords:
Notice what happened. One image held three products, and Generative AI Image Search gave each its own keywords, down to the gum sole.
Generative AI Image Search opens the results for the first keyword. Above them, the image sits beside Analyzed Keywords, each one tap away.

Here the buyer tapped “medium wash denim”. The page shows “These are results for: medium wash denim” with jeans for women, men and kids.
The results page works like any search. Price, color, and other filters, sorting, and grid or list view all stay in place.
For buyers, visual product discovery makes finding a product as easy as showing it:
For merchants, visual product discovery ends more searches on a product page instead of an empty one:
Turning on AI-powered visual product discovery takes a few minutes in the admin panel, with no coding. There are three screens to visit.
Step 1: Switch Magic AI on. Go to Configure → Magic AI → General and turn on Enabled. This one switch controls Magic AI store-wide.

Step 2: Add an AI provider. Under Configure → Magic AI → Providers, paste the API key from your AI provider and save. Bagisto lists eight:
You only need one. Ollama keeps every image on your own server, which suits stores with strict privacy rules.

Step 3: Choose the model. Open Configure → Magic AI → Storefront Features. Under AI Image Search, turn on Enabled and pick a model.
The list shows every model by provider, such as “Anthropic: Claude Opus 5” or “Gemini: Gemini 3.8 Flash”. Pick one that can read images.

The camera icon itself has a switch under Configure → Catalog → Products → Settings → Image Search Option. It is on by default.
Buyers can upload JPEG, PNG, GIF, WebP and SVG images up to 2 MB. HEIC images, the default on many iPhones, are not accepted.
Works with Bagisto extensions:
Visual product discovery itself is built into Bagisto. Your AI provider bills you for each image it reads, at its own rates.
TensorFlow image matching in the browser is free, and so is a model you run yourself with Ollama.
The store saves each image on its own server, so it can show it beside the keywords. It also sends the image to your chosen AI provider to be read.
With Ollama on your own server, the image never leaves your infrastructure.
Yes. The camera icon sits in the mobile header too. Images must be JPEG, PNG, GIF, WebP or SVG, up to 2 MB; HEIC images are not accepted.
The search still works. The store switches to TensorFlow image matching in the buyer’s browser and opens results for the label it finds.
Any model in the list that can read images will work. The choice is a trade-off between accuracy and cost.
For the most exact keywords, pick a flagship model with strong vision, such as Claude Opus 5 or Gemini 3.1 Pro. It costs more per image.
In return, it catches the fine details that make a match: the wash of the denim, the fabric, the cut and a small brand logo.
Lighter models such as the Flash family answer faster and cost less per image. They suit large catalogs of simple products.
Yes. Each channel has its own AI Image Search switch and model. To hide the camera icon itself, turn off Image Search Option.
Product discovery has moved from categories to keywords to images. Visual product discovery lets a buyer show your store what they want.
Your store answers with the right products, whether it reads the image with TensorFlow or with AI.
The same Storefront Features screen holds two more Magic AI tools: Review Translation and Personalized Checkout Message.
Pair it with AI product content generation, which writes the listings those keywords find.
Get started today: Follow the AI Image Search guide.
If you have more details or questions, you can reply to the received confirmation email.
Back to Home
Be the first to comment.