How to Make Marketplace Product Photos with AI: 6 Steps

Making marketplace product photos with AI takes one phone shot: you upload it as a reference, describe the scene, and get a hero image on a clean background, a lifestyle shot, a close-up and an infographic — no studio, no photographer. The model that keeps your product intact best is Nano Banana Pro (shape, color and label stay as in the source), readable text for infographics comes from GPT Image 2.5 Sunburst, and a scene assembled from several references is the job of Seedream 5.0 Pro. All three run in the IIshki studio, paid with tokens. Below: which shots a listing on Amazon, Etsy or Wildberries needs, a step-by-step route, and prompts that don't redraw your product.

Which AI model to choose for product photos

There is no single model for everything: a listing is built from different shot types, and each has its own best tool.

If in doubt, start with Nano Banana Pro: it covers the hero shot and lifestyle, and you switch to GPT Image 2.5 for the infographic later. A detailed comparison of the two is in GPT Image 2 vs Nano Banana Pro, and Seedream vs Nano Banana in Seedream 5 vs Nano Banana Pro.

Which shots a product listing needs

Buyers swipe through a listing rather than study one image. A working set is five to six shots:

1. Hero studio shot — product centered, clean solid background, even light. This is the image for moderation and search results.
2. Product in context — cream on a bathroom shelf, sneakers on a treadmill, a thermos on a campsite table. These show how the item lives in daily use and give the most noticeable conversion lift.
3. Texture close-up — stitching on fabric, matte plastic, a highlight on glass. Answers "what does it feel like".
4. Infographic — headline, three main benefits, size or volume. Often placed second or even first.
5. In hand or on a model — scale and the UGC feel of "a friend took this".
6. Top-down flat lay for bundles — every item in the kit visible in one frame.

All six come from one source image: it goes into every generation as a reference, only the scene description changes.

How to make marketplace product photos with AI: 6 steps

1. Shoot or prepare the source image

The AI transfers what it sees in the reference, so the source matters more than the prompt:

• one product in frame, in focus, no motion blur;
• diffused light without harsh highlights or deep shadows — a window with a sheer curtain beats a ceiling lamp;
• a solid surface under the product: a sheet of paper, a table, a windowsill;
• label and logo facing the camera and legible;
• the original file, not a copy compressed by a messenger.

If the shot is blurry or small, run it through Topaz Image Upscale first at 2× with Text Refine mode — the packaging text gets sharper and the model has something to carry over.

2. Make the hero shot on a clean background

Open Nano Banana Pro, upload the source as a reference, pick 3:4 (vertical listing) and 2K quality.

Take the product from the reference and keep its shape, color, proportions and packaging text unchanged. Place it centered on a clean white background, soft diffused studio light, a light natural shadow under the object. Catalog product photography, no extra objects.

Compare the result with the real product: the cap, the number of stripes on the label, the shade. A mismatch means regenerate, not publish.

3. Move the product into a scene

Same model, same reference, different description. Pick the scene for the buyer, not for "prettier":

Keep the product from the reference unchanged. Place it on a light marble bathroom shelf, next to a folded white towel and a sprig of eucalyptus, soft morning light from a window on the left, shallow depth of field, slightly blurred background. Photorealistic, 3:4.

For an in-hand shot, describe the hand separately — otherwise you get extra fingers:

Keep the product from the reference. A realistic female hand with five fingers holds the bottle at chest level, sunlight, blurred park background, shot like a smartphone photo, light grain.

Need a scene from several samples — your own backdrop, your props, your product? That's a job for Seedream 5.0 Pro: up to 10 references, each addressed by number.

Product from @Image1 on the table from @Image2, a cup from @Image3 next to it. Do not change the product, lighting as in @Image2, photorealistic.

4. Add a texture close-up

Name the material literally — "matte plastic", "polished steel", "dense cotton with visible stitching". A vague "high quality" gets interpreted however the model likes.

Keep the product from the reference. Macro shot of the upper third of the bottle: condensation drops on glass, a crisp highlight along the edge, dark neutral background, hard side light, 2K.

5. Build the infographic

Switch to GPT Image 2.5 Sunburst: the GPT Image series has the most stable text on images. Upload the same source as a reference and list the text in quotes — the model reproduces it as written.

3:4 marketplace product card. Product from the reference on the right, unchanged. On the left, a large headline "Thermo Mug 500 ml", below it three lines with icons: "Keeps heat 12 hours", "Leak-proof", "Fits a cup holder". A badge in the top corner "Bestseller". Light background, readable font, plenty of white space.

Fine regulatory text (ingredients, barcode, certificates) is still weak for AI — add it in a graphics editor on top of the finished image.

6. Check the shots and assemble the set

Before uploading, run every image through three questions: the product matches the real one, the text reads without mush, the shadow falls away from the light source. Finished images sit in "My generations" and download at full resolution from there.

If you have dozens of listings, repeating the steps by hand is slow. On the Canvas, one node with the source connects to several generation nodes — each with its own scene prompt — and the whole set runs at once.

How to write a prompt that keeps the product yours

A product-photo prompt differs from a regular one: what matters is not the beauty of the scene but the ban on changing the product. Five parts worth keeping in every request:

1. Lock the product — "keep the shape, color, proportions and text from the reference". Without it, the model treats the reference as a hint and redraws.
2. Scene — where the product stands, what's next to it, what surface.
3. Light in a photographer's words — "soft morning light from the left", "diffused, no shadows", "hard side light". The most controllable parameter.
4. Material — the surface texture literally: glass, matte plastic, chrome, linen.
5. Format and quality — 3:4 or 1:1 for a listing, 16:9 for a banner; 2K for a scene, 4K if the image has small text.

One task — one prompt. Cleaning the source, the scene and the infographic are three separate generations, not one request for everything. More reference-editing tricks are in Nano Banana Pro prompts.

Common mistakes

What's next

A finished listing is a reason to go further: the static hero image becomes a short clip with a slow product turn in Kling 3.0 — upload the frame as the start image, write "the camera slowly orbits the object, lighting unchanged", pick 5 seconds. The step-by-step is in how to animate a photo with AI.

If you need to process photos you've already shot — retouching, background swaps, removing clutter — see the AI photo editing roundup. And all models for product listings in one list are on the AI for marketplaces page.