How to create marketing visuals for e-commerce: a complete guide
TL;DR: A product listing sells in place of a salesperson: the buyer cannot pick the item up, so the decision is made from images. A working set is a clean product shot, a lifestyle scene, a card with key features, and a short video. Below: where to start, which visuals you need, what technical requirements to meet, how to build the set from one photo, and the mistakes that most often eat conversion.
Key takeaways
- One photo on a white background is not enough — each image type closes a different stage of the buying decision.
- The main marketplace image must be clean; a card with text on it will fail moderation there.
- Technical requirements (format, ratio, file weight) matter as much as the creative: heavy files slow the page and hurt ranking.
- AI image generation for e-commerce builds the whole set from one source photo in seconds, removing the dependency on shoots and designer queues.
- The costliest mistakes are a background competing with the product, text overload, and an inconsistent catalogue.
Why visuals decide the sale
Offline, a buyer picks the item up: they judge weight, texture, stitching, the real shade. Online that channel is gone — everything that would have arrived through touch has to come through the image. That is why a product listing works not as an illustration to the description but as the main carrier of the argument.
The consequence is simple: weak visuals read as a weak product. Buyers rarely put it into words — they just keep scrolling. Conversely, a set that shows the item from the angles that matter, in a real context, with answers to the usual questions, removes doubt before it hardens into a no.
The traditional way to get that set is a photographer, a studio, props and post-production. That works while you have dozens of items. Once there are thousands and the range turns over every season, the cycle breaks: the shoot becomes the bottleneck that keeps products out of the storefront for months. This is why AI image generation for e-commerce today solves a logistical problem rather than a design one — it removes the dependency on a shooting schedule.
Where to start
Audit the current process
Before changing tools, work out what one visual costs today — not only in money but in the time from "the product arrived at the warehouse" to "the listing is live". Usually the shoot itself is a small slice, and the bulk goes on approvals, retouching and preparing formats for different platforms. Those are the stages to automate first; the payback is fastest there.
Brand guidelines
Even with automation you need written rules: palette, standard angles, text density, the character of the background. Without them the catalogue drifts — neighbouring listings look like they came from different stores. Guidelines save more than designer time: they turn a vague "make it look good" into something you can actually check.
Pilot on one category
Do not redo the whole catalogue at once. Take one category, build the full set, publish it, then look at the metrics and at how marketplace moderation reacts. A pilot exposes the real bottlenecks in your process rather than the hypothetical ones, and it costs little if something goes wrong.
Which visuals a product listing needs
The clean product shot
The catalogue baseline. The product on a plain, even background with nothing else in frame, so the buyer can judge shape, colour and proportion objectively. This is the standard for the main marketplace slot and the core requirement of their moderation.
The lifestyle scene
The product in a real environment or in use. Its job is not to show but to let the buyer imagine: how the thing looks at home, on a person, at work. For apparel, furniture and decor, lifestyle outsells studio — there, the context is the argument.
The feature card
A full layout: headline, key benefits, size, price directly on the image. It answers questions before the buyer goes looking for them. Its place is the additional gallery slots, catalogue banners, ads and social.
Video
The card brought to life: the product turns, the text stays put. Video holds attention longer than a still and answers "what does it look like from the other side". In most generative models text drifts and breaks apart once motion starts, which is why captions usually get dropped; when the text holds pixel for pixel, the card keeps its meaning in motion.
How it fits together. The main photo is clean and wins attention in the listing grid. Then a few angles and details. Then lifestyle for emotional pull. Then the feature card. Video goes into whichever slot supports it, or into ads. That way the buyer gets everything through the visual channel, without reading the description.
Technical requirements
Formats and file weight
Resolution has to be high enough for on-site zoom to show texture and stitching. But large files slow the page, and load speed affects both user behaviour and ranking. The compromise is modern compression: WebP gives better detail at lower weight than JPEG or PNG.
Prepare aspect ratios for the channel up front: 1:1 for the catalogue, 4:5 and 3:4 for feeds, 9:16 for stories, 16:9 for banners. Re-cropping by hand afterwards is exactly the manual step you were trying to remove.
What marketplaces require
Rules differ across platforms, but the principle is shared: the main image must be clean — no text, badges or promotional elements. A marketing card with text in the primary slot will simply fail moderation. Ukrainian platforms — Rozetka, Prom.ua, Horoshop, stores on OpenCart — apply this as strictly as the international ones.
The additional gallery images, catalogue banners, assets for Kasta and Allo, ads and social posts are where a feature card does the conversion work. Plan the set around that split from the start, or half the material will need redoing.
Mobile readability
Most purchases now happen on a phone. Text on a card has to be readable without zooming: if a spec is only legible on desktop, it effectively does not exist. Check the card at real size in a feed, not on a large screen in the editor.
Building the set, step by step
Step 1: the source photo
One photo with even lighting, showing the product from the angle you want. A smartphone shot works — what matters is adequate resolution and no harsh highlights eating the texture. The system identifies the product and separates it from the background, so there is no manual cutting out.
The uploaded product stays in the library, so later generations do not need it again. For a large catalogue that saves more time than the generation itself.
Step 2: asset type and scene
Next you choose what to get: a clean product shot, a lifestyle scene, or a card with text. Here you also set the scene, the product placement (right, left, centre) and the aspect ratio for the channel.
Photos and cards come back in seconds, so variants can be explored rather than planned in advance. More on the AI generation page.
Step 3: video
Video is built from a card: either animate an existing one, or assemble a card from the photo first and then animate it (the two stages are billed separately). You set duration — five or ten seconds — quality, and looping. The finished clip is MP4.
Video generation takes one to five minutes, longer than a still. If the chosen model's moderation rejects the frame, the credits are returned.
What changes by category
The most common mistake is treating every product the same way. Presentation rules differ, and this is where AI image generation for e-commerce beats generic templates: composition is matched to the category.
Apparel and footwear. Fit and fabric texture carry the sale. Either show the item in full or crop on a clear line — a crop landing somewhere in the middle reads as a defect. Season shows visually too: a winter piece in a summer scene undermines trust.
Cosmetics. The surface sells: the gloss of a bottle, the matte of a tube, the translucency of a gel. Reflections and colour accuracy are critical here — shade is the single most common reason these products come back.
Food. Appetite and scale do the work: the buyer needs to understand the real portion size. Over-styling hurts here — it sets up an expectation the product cannot meet.
Tech and large items. Dimensions and build quality matter most. These benefit from a scene that conveys size, plus close-ups of the parts buyers actually care about.
What to check after generation
Automation does not remove the final check — it just makes it fast. Look at three things.
Colour. Compare against the physical product, not against your screen. This is the most expensive mistake: a shade mismatch converts directly into returns.
Proportion and shadow. The product should look embedded in the scene: the shadow falls in a direction consistent with the lighting, and the scale matches its surroundings.
Text. Readability at real size, no typos, and figures (price, size, contents) that match the actual listing data.
General-purpose editor or specialised tool
General-purpose design platforms handle the basics well: remove a background, build a collage, drop in a template. But they treat the product as a flat cut-out — the object sits on the backdrop rather than inside it. Shadows are notional, contact shadow and reflections are missing, and the buyer reads that as false within a fraction of a second, even without being able to name what is wrong.
Background-removal tools solve one narrow task and stop there: you still need a designer to turn the cut-out into a finished card.
A specialised tool differs in that the product stays untouched — same shape, same shade, same texture — while the scene, lighting and composition are built around it for that specific category. That is why the card comes out fit for the catalogue rather than merely product-shaped.
There genuinely is no universal tool — but stitching three services together for one set of visuals means restoring the manual process you were escaping.
Common mistakes
A background that competes with the product. A complex environment should support the product, not pull attention away from it. Restraint and a clear focal point consistently win on conversion.
Text overload. Too much copy makes the card chaotic, kills readability on mobile and reduces reach in ads. Keep only what matters on the image: the key benefit or the size of the discount.
An inconsistent catalogue. When neighbouring listings have different lighting, angles and background styles, the store looks unprofessional. Lock the settings for a whole category instead of picking them again each time.
Ignoring the platform's rules. The most frequent rejection is a designed card in the main slot. That is a formal rule, not a moderator's taste.
Quality control and testing
Keep a pre-publish checklist: shadow realism, product proportions relative to the scene, colour accuracy, text readability at small sizes, and compliance with the platform's rules. It standardises the output and removes the human factor.
Then run A/B tests. Produce alternative scenes for the same product and push them with a small budget, watching CTR and conversion. When a variant costs seconds, testing becomes routine rather than quarterly.
Finally, look at why products come back. Complaints about colour or size not matching the photo are a direct signal that something is wrong with the visuals — and the cheapest way to learn it before the metrics drop.
FAQ
How many images does one product listing need?
A working minimum is the main product shot, two or three angles or details, one lifestyle scene, and a feature card. Video is added wherever the platform or channel supports it.
Why is the marketplace rejecting my image?
The most common cause is text, badges or promotional elements in the main slot. That slot needs a clean product shot; anything designed belongs in the additional gallery images.
Which format and aspect ratio should I use?
WebP on the site as the quality-to-weight compromise; ratios by channel — 1:1 for the catalogue, 4:5 and 3:4 for feeds, 9:16 for stories, 16:9 for banners. Produce the ratio you need up front instead of cropping later.
Do I still need a designer if I use AI?
Not for assembling the listing. A designer stays valuable for brand guidelines and non-standard campaigns, but routine catalogue updates no longer depend on one.
Will a smartphone photo work as the source?
Yes, provided the lighting is even, the resolution is adequate, and there are no harsh highlights on the product. A studio source frame is not required — a legible product in frame is.
Conclusion
Good visuals in e-commerce are not a matter of taste — they are part of the sales funnel. A set of clean product shot, lifestyle scene, feature card and video covers the buyer's decision end to end, and AI image generation for e-commerce from a single photo makes that set realistic even for a catalogue of a thousand items.
More on how AI is changing the economics of product photography in How AI is changing product photography. Plans and terms are on the Pricing page.
Ready to try it on your own product — start for free: new accounts get starter credits, no card required.