The catalogue asks for one honest image of each product. The campaign asks for the same product in a scene, in a square and a vertical and a story, with room for a headline, in a spring palette this month and a gift palette in November, for the feed and for the paid placement and for the email, and it asks again next month. Nobody photographs that; it is produced. The question is how to produce it without the product drifting away from the thing that ships.
This page is a method for making social and paid-campaign product images with AI. It publishes no figure for engagement, click rates or cost per image, and it does not describe any platform's advertising policy, which belongs on the platform's own page. What it gives is a way to build the campaign set from the masters the catalogue already has, the line between a creative scene and a false claim, and a checklist per placement.
How do you create social media and ad images for products with AI?
Start from the catalogue's masters — one sharp photograph per product on a plain ground — never from a description of the product. Decide the set per campaign: the placements, the formats each needs, the message and the palette. Generate the scene around the kept product for each format, at the framing that placement shows, leaving the space the copy will occupy; for concept visuals that are not the product — a mood, a setting, a texture — use a prompt-driven canvas and keep the product out of it or composite the kept product in. Review every image against the master before it goes into a campaign, and keep a single file naming and crop rule so the set can be rebuilt next month by changing only the scene.
What a campaign image is, and how it differs from a listing image
A listing image describes the product to someone deciding whether to buy it, on a neutral ground, at a channel's fixed size. A campaign image carries a message to someone who was not looking: a season, a mood, a price, an occasion. It lives in a placement with its own format and its own crop, sits beside copy, and is replaced on a calendar. The product's pixels are the same in both; what changes is everything around them. The failure specific to campaigns is a scene so persuasive that the product inside it stops being the product.
The set, decided before anything is generated
- Placements: feed, story, reel cover, paid placement, email header, landing hero. Each has a format and a safe area.
- Formats: square, vertical, wide; the exact ratios come from each platform's current specifications.
- Message and palette: one per campaign, written down, so every image in the set reads as one campaign.
- Copy space: where the headline sits in each format, so the scene leaves it empty.
- The product's role: hero, prop, or absent; a concept visual with no product needs no product check, and an image with the product needs the whole one.
Make the set from the masters
- Kept product, generated scene. For every image that shows the product, the product's pixels come from the master; the scene, the palette and the props are generated around it. In Photta the lifestyle scenes and the background changer work this way.
- Concept visuals without the product. A mood image, a texture, a setting: made from a prompt and reference photographs in Photta Canvas, with the product either absent or composited in from its master.
- One scene per format, not one crop per scene. A square and a vertical of the same scene are generated at their own framings so the product sits where the placement shows it, rather than cropped from one wide image.
- Variants for testing. Two palettes or two settings of the same kept product are a paid-media test; two versions of the product are not, and never a test.
- Exports per placement, named by campaign, product, format and variant.
Where the creative line runs
A campaign scene may be anything the brand wants: a beach, a kitchen, a gift table, a colour the season owns. The product inside it may not be anything but itself. The line runs through the product's pixels:
- Allowed: a scene, a mood, a palette, props, a wearer generated around a kept garment, a season that does not exist yet.
- Not allowed: a feature the product does not have, a finish it does not come in, a size it is not, a result it does not produce, a before-and-after that never happened, text or badges on the image that make a claim the page does not make.
- The test: would the customer who bought from this image find a difference in the box? If yes, the image is a claim, not a scene.
Platform rules on advertising content are the platform's, change often, and are read on the platform's own page before a campaign runs.
Set, scene, review
The loop the checklist below serves, named so the other catalogue pages can point at it.
- Set — placements, formats, message, palette and copy space decided and written before generation.
- Scene — generated around the kept product per format; concept visuals made without the product or with it composited in.
- Review — every image with the product compared to the master; every image checked in its placement's frame with its copy.
Campaign image checklist
| Check | Pass | Fail |
|---|---|---|
| Product pixels | The master's, in every image that shows the product | Redrawn, recoloured, resized, a feature added |
| Scene | Generated around the product; the light agrees with it | The product floating; shadow against the light |
| Format | Generated at the placement's framing | Cropped from a wide image with the product cut |
| Copy space | Empty where the headline sits; readable over it | Copy across the product; a busy area under text |
| Message and palette | One per campaign, the same across the set | Twelve images that look like twelve campaigns |
| Claims | None the product page does not make | A badge, a result, a before-and-after |
| Concept visuals | No product, or the kept product composited in | A product invented by the prompt |
| Variants | Scenes vary; the product does not | Two versions of the product in one test |
| Naming | Campaign, product, format, variant in the name | Files nobody can rebuild next month |
| Platform rules | Read on the platform's page before the run | Assumed from last season |
The claims row is the one a creative team fails without noticing: a persuasive scene invites a persuasive product.
One candle, one season, one set
A candle brand plans an autumn campaign for three scents: a feed square, a vertical story, a paid placement and an email header, each in an amber-and-linen palette with a headline on the left. The masters are the catalogue's — each jar on white with a grey card. For every format the scene is generated around the kept jar at that framing, with the left third empty; the jars' labels are read against the masters on every image. Two concept visuals — a table of dried flowers, a window with autumn light — are made in Canvas with no product in them. One paid variant showed the jar with a wooden lid the product does not have; it was rejected at the claims row and regenerated. The set is named by campaign, scent, format and variant, and in November only the scene and the palette change.
Frequently asked questions
Can I use AI-generated product images in paid ads?
You can use images in which the product's pixels come from your photograph and the scene is generated around it, provided the image makes no claim the product page does not make and follows the platform's current advertising rules, which you read on the platform's own page. An image in which the product itself was generated from a description is a claim about a product that does not exist.
How do I keep the product honest inside a creative scene?
Keep its pixels: generate the scene, the palette and the props around the master, never the product from a prompt. Then ask of every image whether a customer who bought from it would find a difference in the box. If the answer is yes, the scene has become a claim.
Should I crop one wide image for every placement?
No. Generate each format at its own framing so the product sits where the placement shows it and the copy space is empty where the headline goes. A crop from a wide image cuts the product or fills the copy space.
What is Photta Canvas for in a campaign?
Concept visuals that are not the product: a mood, a setting, a texture, made from a prompt and reference photographs, in several variants at once. The product either stays out of those images or is composited in from its master, so its pixels are never drawn by the prompt.
How do I make the same set again next season?
Write the set down — placements, formats, message, palette, copy space — and name the files by campaign, product, format and variant. Next season only the scene and the palette change; the masters, the formats and the names stay, and the set rebuilds from the same recipe.
What this page does not claim
That AI campaign images perform at any rate, that any platform accepts or rejects them, or that any set converts; this page measures none of those. What it claims is that a campaign image is a listing image with a placement and a message, that the product's pixels are the line the scene may not cross, and that a written set is what makes next month cheap.
To make concept visuals and variants from a prompt, see Photta Canvas; for who reviews what in a team that produces sets, building an AI creative operations team; for the images a campaign should photograph instead, when AI product photography is not the right choice.