How to Generate Shopify Product Images in Bulk with AI
Most catalogs aren't short of product photos. They're short of consistent ones. The images arrived with the products — from a supplier, a CSV import, three different vendors and someone's phone — and every one was shot in a different room, on a different surface, at a different distance.
Individually they're fine. Together, on a collection page, they look like a market stall: one product floating on white, the next on a grey desk with a keyboard in shot, a third cropped so tight you can't tell how big it is. The catalog reads secondhand before a shopper has read a word.
Re-shooting is the obvious fix and the reason nothing gets fixed. A studio day costs real money and produces images only for the products you remembered to send. The alternative is to regenerate the product images you already have — every product at once, rather than one at a time.
One supplier photo, three prompts, three new images — the same blankets each time.
What "regenerate" actually means
Regenerating a product image means producing a new version of a photo you already have — same product, new staging. The word matters, because three different things get sold under the same heading and only one of them is safe for a product catalog.
- Generating creates an image from a text prompt alone. The result is a plausible product that has never existed. Fine for a mood board, wrong for a listing.
- Editing changes the pixels you have — cropping, compressing, cutting out a background. Nothing new appears; nothing new can appear.
- Regenerating starts from your photo and produces a new version of it. The product's shape, color and material come from the original. The background, lighting, framing and staging are what the prompt changes.
Everything below is about the third one: how to regenerate product images across a whole catalog rather than one product at a time.
When to regenerate, and when not to
Regeneration isn't a universal answer. It's the right tool when the product is fine and the photograph is the problem.
| Situation | Best move |
|---|---|
| Supplier photo, cluttered background, product clearly visible | Regenerate — a clean background is a prompt away |
| Catalog shot by a dozen vendors in a dozen styles | Regenerate the lot with one prompt, for a single look |
| Good photo, wrong shape for your grid | Reshape — no AI needed, and a tenth of the price |
| Good photo, 4 MB file slowing the page | Compress — same image, smaller file |
| Product has printed text, a logo or a spec label | Careful — AI mangles text; check every result, or leave it |
| Color or finish must be exact (paint, fabric, jewelry) | Photograph it — don't let a model guess a colorway |
What size should Shopify product images be?
Shopify publishes four image recommendations, and one of them — a single aspect ratio — is the one a bulk run is good at. In full: 2048 × 2048 px for square product images, up to 5000 × 5000 px or 25 megapixels, files under 20 MB, PNG first and JPEG second, and — the one most catalogs quietly break — one consistent aspect ratio across your images, so collection pages don't jump around.
Be precise about which of those a regeneration run fixes. Shape, yes: pick square or portrait before the run and everything comes back in it, which is the fix for a grid that crops every card differently. File size, separately: that's compression, at a tenth of the price. Resolution, no. Generated images land around 1,000 to 1,250 px on the long edge — across our own production runs the most common output is 1024 × 1024. That's ample for product cards, galleries and search listings, and short of the 2048 px Shopify suggests if you rely on deep zoom. For a hero image someone will pinch-zoom on a phone, keep the photograph.
Four jobs, four prompts
Most AI product images come out of four prompts, and a prompt that works on one product often falls apart across a catalog because it assumed something only that product had. These four survive the jump — each with the job it does and the way it fails.
1. How to make supplier photos look professional
"Professional photo on a clean white background, soft shadow"
The workhorse, and the biggest single upgrade on an imported catalog: it turns a kitchen-table photo into something deliberate, and it's what marketplaces and product cards expect. Do this one first. Watch for shadows falling in a direction nothing else in the frame agrees with.
2. A lifestyle setting
"Lifestyle product shot on a bright wooden table, morning light"
Answers the question a cut-out never does: what does this look like in a real room, and how big is it? Name a surface and a light source rather than a whole room — the more scenery you ask for, the more the product competes with it. Watch for props that imply something is included when it isn't.
3. A close-up
"Close-up detail shot highlighting texture and craftsmanship"
Swap the last two words for whatever material matters: grain, weave, finish. A detail shot costs a few cents and answers the question a shopper is zooming in to ask; most imported catalogs have nothing to zoom into. Watch for invented texture on surfaces you can't verify in the original.
4. A seasonal restyle
"Styled for Christmas, warm lights and pine branches"
The same catalog, dressed for a campaign. Because originals are backed up you can put the supplier photo back in January — but restoring returns the original, not whatever you published before the campaign. If October's white-background run is the look you want to come back to, keep those images on the products rather than restoring over them.
Notice what those prompts don't do: name the product. The model is looking at your photo, so it already knows what the thing is — the prompt's job is the staging around it. There's a {product_name} placeholder that drops each product's title into its own prompt, and it's worth reaching for when the title carries something the photo can't show (a material, a size, a model number) or when the scene should match the product type. Most runs don't need it.
Two more habits save credits: keep the prompt about staging rather than the product itself, and when a result is close but wrong, change one thing at a time.
Running it in bulk across a catalog
Run it in this order and you spend credits on images you keep.
- Start with one collection, not the catalog. Filter to twenty or thirty products that share a problem — one vendor, one import batch — and work out the prompt there.
- Write the prompt in plain language. Describe the staging, not the product — the model can see the product in the photo you gave it.
- Pick the output shape before you run. One choice applies to the whole batch: square for product cards, portrait for fashion.
- Keep review on. Results should wait where you can see them next to the originals, not land on live product pages.
- Publish the good ones, discard the rest. Expect to discard some — a prompt that works on nine products in ten is a good prompt. Discards still cost their credits, which is exactly why step one is a single collection.
- Then widen the scope. With wording that works, run the same prompt on the collection, the vendor, or everything.
- Changed your mind later? Restore from backup on the finished run puts the originals back, filename and alt text included — publishing isn't the point of no return.
Every result sits beside the photo it came from, waiting for a decision.
What to check before you publish
Check every result for:
- Garbled text. The most common failure by a wide margin. Logos, care labels and packaging copy come back as convincing nonsense — letterforms that look like words until you read them. Any product whose photo shows text needs a close look.
- A product that drifted. Handles that changed shape, a different number of items in a set, a strap that grew. Compare against the original, not against your memory of the product.
- Physics that don't match. Shadows falling the wrong way, a reflection with nothing to reflect, an item floating a centimeter above the surface.
- Accidental promises. Props that suggest something is included when it isn't. If the box doesn't contain the styled bowl of coffee beans, the shopper who thinks it does will tell you in a return.
- Consistency with the rest. The new image has to sit next to your existing ones without looking like it came from a different store.
A glance per image, next to the original — ten minutes for a hundred results. The review step is the difference between a catalog that looks professionally shot and one that looks convincingly wrong.
What it costs
Here's the arithmetic. With Norlif AI Image Generator a generated image costs 10 credits, and credits come in one-time packs: $5 for 500, $26 for 4,000, $56 for 10,000. At the middle pack that's 6.5 cents an image. Reshaping or compressing an existing image costs 1 credit — a tenth of that.
One new image for a 400-product catalog is 4,000 credits: $26, and one confirmation. Three looks for each of those products — white background, lifestyle, close-up — is 12,000 credits, about $82 in packs. A studio day starts at several times that, and only covers the products you remembered to ship.
Budget for waste, because discards aren't refunded: credits are spent when an image is generated, not when it's published. At a 90 percent keep rate that 400-image run loses about 400 credits — roughly $2.60. Testing the wording on thirty products first costs 300 credits and usually saves more than it spends.
Installing is free and includes 100 credits — ten generated images, enough to find out whether your prompt works on your products before you spend anything. Install Norlif AI Image Generator from the Shopify App Store and run it on one collection first.
The short version
Mixed, inherited product photography is a catalog-wide problem. Regenerating starts from the photos you already have, so the product stays yours while the staging changes. Work out the prompt on one collection, pick a single shape so your grid stops jumping, review every result against its original — watching for mangled text above all — and only then run it across everything. At a few cents an image, the expensive part isn't the photography any more. It's deciding what you want the catalog to look like.
Frequently asked questions
What does it mean to regenerate a product image?
Regenerating means producing a new version of a photo you already have: the same product, restaged. You give the AI your existing image plus a prompt — a clean white background, a lifestyle setting, a close-up — and it returns a new image built from that photo. That's different from generating an image from nothing, which invents a product that never existed, and different from editing, which crops or compresses the pixels you already have.
Will the product still look like my product?
Mostly, and that's exactly what you have to check. Regeneration keeps the shape, color and material of the original because it works from your photo — but it can drift on fine detail, and it's unreliable with text. Logos, care labels and packaging copy are the first things to go wrong. Review every result next to its original and discard anything where the product itself changed.
Which image does it regenerate?
The product's main image — one new image per prompt, per product. Write three prompts and each product comes back with three new versions of its main photo. You choose whether the new image replaces the original on the product or sits alongside it, and either way the original is backed up first.
What image size does Shopify recommend?
Shopify recommends 2048 x 2048 px for square product images and accepts up to 5000 x 5000 px or 25 megapixels, with files under 20 MB. PNG is its first choice, then JPEG, and WebP is accepted. It also recommends one consistent aspect ratio across your images so collection pages line up. Worth knowing: AI-generated images come back around 1,000 to 1,250 px on the long edge, which is fine for product cards and galleries but below the 2048 px Shopify suggests for deep zoom.
Can I regenerate images for a whole catalog at once?
Yes, and past about twenty products it's the only sane way to do it. Pick a collection, a vendor or the entire catalog, write one prompt, and every selected product gets its own new image built from its own photo. A 400-product run is one confirmation rather than 400 sessions in an editor.
What does it cost per image?
With Norlif AI Image Generator a generated image costs 10 credits. Credits come in one-time packs — $5 for 500, $26 for 4,000, $56 for 10,000 — so a generated image works out between 5.6 and 10 cents depending on the pack. Reshaping or compressing an existing image costs 1 credit. Installing is free and includes 100 credits, which is ten generated images.
Do I pay for images I discard?
Yes. Credits are spent when an image is generated, not when it's published, so a result you discard still cost its 10 credits. That's the argument for testing a prompt on a handful of products before running it across a catalog — at a 90 percent keep rate, a 400-image run wastes about 400 credits, or roughly $2.60.
Does it write alt text for the new images?
The published image inherits the alt text of the photo it came from, so nothing is lost — but it doesn't write fresh alt text describing the new scene. If the staging changed enough to matter for accessibility or image search, update the alt text yourself.
What if I run the same product twice?
It generates again and charges again. There's no skip-if-already-done for generation the way there is for compression, so a second run on the same collection is a second bill. Filter to what actually needs redoing before you run.
Does Google penalize AI-generated product images?
Google's stated position on AI content is about quality and intent rather than how something was produced — it rewards helpful, accurate content and acts against content made to manipulate rankings. The real risk with product images isn't the algorithm but the shopper: an image that misrepresents what arrives in the box drives returns and chargebacks, whoever or whatever made it.
How long does a bulk run take?
About four images a minute, which is the median across our own production runs — so a hundred images take roughly 25 minutes and a four-hundred-product catalog runs for an hour and a half. You don't wait around for it: runs process in the background, results appear as they finish, and you can close the app and come back.
Images are one of several fields an imported catalog leaves in poor shape. Read next: why your product titles are too long, how to write product descriptions in bulk, and how to auto-assign Shopify product categories with AI.