AI Prompts for Product Listing Photos: Build the Slot Map First

A channel does not judge your product photo on taste. It checks a background, a fill share and a minimum size - and then a shopper asks six separate questions. Fix the slot map before you write the AI prompts for product photos, and the set stops being guesswork.

The main image of a listing is one field in a feed, not the whole job. Every additional image is a slot with its own question attached, and some of those questions cannot be answered by any model: the answer lives in your inventory or your spec sheet. Naming that boundary before you prompt is the skill. The model puts the scene, you put the facts.

AI Prompts for Product Photos: The Slot Map

A slot map is a short table built once per product category: slot number, the question the shopper is asking, the field that answers it, the rule it must satisfy, and who produces it - model, records or a real camera. Six rows, five columns, twenty minutes. It saves a week of re-uploads, because you stop discovering the rules after the photo exists.

The feed already tells you the set is structured. In Google Merchant Center the main image is submitted in one attribute, alternative angles and detail shots go in another, and a third is reserved for an in-context or lifestyle representation. The image link attribute reference names each field. That is the shape your design should follow: one primary field, then a queue.

Six Slots, Six Questions

• Slot 1 - What is it. Product alone, clean field, no props. Recognition, not desire.
• Slot 2 - How does it look from behind or inside. The side a shopper never sees in a catalogue but always turns over in a store.
• Slot 3 - What does the surface feel like. Close detail of texture, seam or grain. One material decision per frame.
• Slot 4 - How is it used and how big is it. In-use scene with a person or a known reference object for scale.
• Slot 5 - What comes in the box. Everything the buyer receives, laid out and countable.
• Slot 6 - Which variant is this. The colour, size or model that separates this listing from the next one.

One slot, one question. An image that tries to answer two answers neither: a hero shot with a size chart burned into the corner is not a main image and not a specification sheet. It is a rejection.

The Rule Sheet Is Not the Prompt

Copy the channel's published rules before writing a word of prompt, then write the prompt against that sheet. Google Merchant Center, for its own listings, recommends a clean, solid white or light neutral background and states that the product should fill no less than 75% and no more than 90% of the frame. Its image best practices page also asks for a crisp hero angle, recommends 1500 x 1500 pixels or larger, notes a 500 x 500 minimum for all products from 31 January 2027, and lists overlaid text, watermarks, promotional claims and added logos as disapproval reasons.

Those numbers belong to one channel. Another marketplace publishes its own, and a third has none beyond a file size. So the rule sheet is a per-channel document, not a memory: save the current page, date it, re-read it when a listing is rejected. A good-looking prompt that breaks the background rule is a rejection with nice lighting.

Four Slots the Model Renders Well

Slots one to four are visual decisions: light, angle, framing, material. When the decision is written down, a model renders it well, because nothing is left to invent except pixels. Slot one names the product, the field, the light direction and the share of the frame. Slot two names which side. Slot three names the material and how close. Slot four names the scene, the scale reference and whether a person appears.

What changes between those prompts is the field you fill, not the model. That is the practical argument for the map: one skeleton with four variants instead of four unrelated paragraphs. Keep a real camera for slot one whenever the product carries printed information, because a model rebuilds a label from a description and a rebuilt label is never the label the buyer receives.

Two Slots That Come From Records

Slot five, and slot six in its measurement form, do not come from a prompt. The model does not know how many cables are in the box, whether the manual is printed, or what the product weighs. Ask for a what-is-included layout or a size table and it invents, confidently and in perfect light. That invented table is worse than no table, because it looks like data.

Those two slots come from a different pipeline: inventory and the spec sheet. You write the list, check it against a physical unit, and compose the image with the list beside the product instead of generating around it. Measurements come from a ruler; the picture only shows where the ruler was placed. It is the cheapest quality control in the set and the most often skipped, because generating feels like progress and counting boxes feels like admin.

The Main Slot: A Background Is a Value, Not a Colour

Look at how a prompt encodes its own rules. MJ Product Ad Backgrounds is a good specimen: a backdrop plate, soft gradient in a chosen colour, subtle paper texture, light from the upper left, empty space in the centre for the product, elegant shadow, portrait ratio. Read as a rule sheet it teaches two things. It already forbids text, which is the right instinct. And it reserves the centre for a product placed later, which is exactly what a main listing slot does not want, because the channel asks for a product filling the frame, not floating in a decorated void.

That does not make the prompt bad. It makes it an advertising piece. Behind a composited main image it is excellent; submitted as the main image it is a background with nothing to recognise. Read every prompt as constraints first and as a scene second, then ask whose constraints they are.

The Slot Your Own Site Allows and the Channel Does Not

FLUX Advertising Campaign Shot goes the other way: concept, staging, dramatic light, colour grade, shallow depth, cinematic composition with negative space for a headline, no text, no logo. An image with freedom - and that freedom is what a slot with a fill rule does not admit as a primary image.

Two images of the same product can both be correct and only one admissible. The difference is not quality, it is jurisdiction. The campaign shot belongs on your site, in the ad, in the social post, where the brief is attention. The listing belongs to the channel, where the brief is recognition and comparability. Some channels accept a lifestyle image in a secondary slot, and Google publishes an attribute for exactly that, but a main slot is a different argument. Sort by jurisdiction before sorting by how good the photo looks.

The Check Before You Upload

Three checks, no tools. First, the slot count: how many images the channel accepts and how many you have, with each question written beside its row. Empty rows are the honest part. Second, the rule sheet: compare the image against the published rules, not your taste - background, fill share, minimum size, no added text. Third, the buyer test: cover the description and look only at the images. If you cannot answer what it is, how big it is and what is in the box, the set is incomplete even when every frame is beautiful.

Frequently Asked Questions

Can AI generate the main image for a marketplace listing? Where the channel's rules allow it, yes - the scene is generated. The label text, the measurements and the box contents are not, because they come from the product and your records.

Why does my generated product look slightly wrong next to the real one? The model rebuilds the object from a description, so proportions, closures and logos drift. For printed detail, the real photograph keeps the main slot.

Do I need all six slots on every channel? No. The channel decides. Below its minimum the set is cut from the tail, so protect slots one and two first.

Start With the Table, Not the Prompt

Build the six rows, one question each, and mark the two your records answer instead of a model. Then write prompts for the four that remain - the background plate for a clean field, the campaign scene where attention is the brief - and check each against the channel's published sheet. The model puts the scene. The rules and the inventory come from whoever has the product in their hands.