Meta catalog ads with AI: turn your Shopify feed into catalog creative

Meta catalog ads with AI creative work like this: you generate better images or short videos for each product, host them, and point the image and video fields in your product feed at them. Meta's catalog ads then serve that creative automatically. For a Shopify store, the hard part is doing it for every product, inside Meta's feed specs, without hand-editing 200 rows.

This playbook covers what Meta's own AI already does, the feed rules your creative has to fit, how TikTok differs, and a step-by-step run from a Shopify catalog, with real per-product costs.

What are Meta's Advantage+ catalog ads?

Meta describes Advantage+ catalog ads as ads that "automatically show people the most relevant products or services from your catalog based on their interests, intent and actions online." They run as single image, carousel or collection ads. The developer FAQ notes they were "previously called dynamic ads."

The creative comes from your catalog. By default that means your product photos: usually a packshot on white, which is accurate but looks like every other store's carousel.

What does Meta's own AI already do for catalog ads?

More than many teams realize. The same help page says Meta can "automatically show people either product images or videos based on what they're likely to engage with" and "apply Advantage+ creative enhancements for catalog such as dynamic overlays, image touch-ups, standard labels and image background generation."

So Meta will already generate backgrounds behind your packshots. What it won't do is follow your brand's art direction, show the product in use, put it on a person, or make a video where you only supplied a photo. That's where your own AI creative comes in. If you want the wider trade-off between letting the platforms generate and running your own pipeline, we covered it in Meta, Google and TikTok will generate your ads now.

Which feed rules does AI creative have to fit?

Your new creative lives in the feed, so it has to pass the feed's checks. From Meta's catalog field reference and help center, checked October 6, 2026:

Two practical consequences. First, keep one aspect ratio across the set: Meta suggests 1:1 or 4:5 so carousels aren't adjusted automatically. Second, the image URL must be stable. If your AI images live at links that expire, the next scheduled fetch breaks the ad.

How is TikTok's catalog different?

TikTok's catalog parameters require nine fields too, but the product ID is sku_id, not id. Square images for carousel ads should be at least 500x500 and end in JPG or PNG. The optional video_link field recommends vertical 9:16 video at 720x1280 or larger, which is a different shape from Meta's 4:5 recommendation. Feeds can be CSV, XML (RSS or Atom), ZIP or GZ (help).

If you serve both channels, plan for two video crops per product, or one vertical master you reframe.

Where does Shopify fit?

Shopify's Facebook & Instagram channel syncs your products to a Meta catalog and is free on all Shopify plans, subject to eligibility. That sync carries your Shopify product photos. To use AI creative instead, you either replace the images in Shopify (which changes your store) or run a separate feed for ads that keeps your store photos untouched. The second option is the cleaner one, and it's the one below.

The playbook: AI catalog creative from a Shopify store

We'll use Orisu's Feed, which is our product. The steps are the same in any setup; the Feed just keeps them in one grid.

1. Import the catalog

Upload a CSV export or connect Shopify. Orisu proposes a field mapping you confirm once, and later re-imports only touch rows that changed.

2. Clean the data

Titles over 65 characters, missing brands and prices in the wrong format will trip Meta's checks later. Ask the Feed to normalize a column and it writes a small script, shows you the before and after on sample rows, and saves the script to replay on every future import.

3. Mount a creative agent as a column

Pick an agent, for example "lifestyle shot of this product in a kitchen, on-brand, 1:1," or a community app. Orisu matches the agent's inputs to your columns (product image, title, category). The agent reads your Brain, so the brand rules apply to every row. If your packaging text must stay exact, use a workflow that composites the real product photo rather than redrawing it; we explain why in AI product ads that keep the real label.

4. Test one row, then estimate

Run the agent on one real product first. When the result is right, run the column. Batches above your cost threshold show the estimate and wait for your approval.

5. Review in one drawer

Each finished cell opens in a review drawer: approve, reject or re-run, then jump to the next one. Only approved rows are eligible for export.

6. Export a feed Meta and TikTok pull

Pick Meta or TikTok, confirm the mapping, and Orisu hosts a CSV at a stable URL. Paste it once into Commerce Manager or TikTok Catalog Manager as a scheduled feed. Every mapped field is checked against that channel's catalog spec before a row goes out: a row that fails a required check is blocked with the reason, and an over-long title is truncated and flagged.

7. Re-run when the catalog changes

New products come in as new rows on the next import, and you run the column for them. Cells whose inputs didn't change aren't charged when you re-run the column.

What does AI creative for 200 products cost?

List prices from Orisu's model catalog on October 6, 2026, one output per product, with dollars at the Pro plan rate ($49 for 5,000 credits, $0.0098 a credit). Retries are not included.

A full image set fits inside a Pro month. The video set needs more credits than Pro includes, so it fits Enterprise (25,000 a month) or a top-up. Rejected cells that you re-run cost again, so test the agent on a few rows before running the whole column. Per-second rates for every video model are in the AI video model price index.

Common mistakes

• Changing the product. A generated bottle with a slightly different label is a returns problem, not just a creative one. Composite the real product.
• Mixed aspect ratios. A carousel of 1:1, 4:5 and 16:9 images gets resized by Meta and looks uneven.
• Image links that expire. The next scheduled fetch fails and the ad stops serving.
• Skipping review. One bad image in a catalog of 200 still reaches real shoppers.

Start with ten rows

Import ten products, mount one agent, and look at the results before you run the rest. The Feed page shows the full grid, ecommerce setups are covered on the ecommerce use case page, and the DTC ads and product ads workflows are good first agents. Credits and plans are on the pricing page.