GPT Image 2.5 Product Photo Prompts: 10 E-commerce Shots

A product listing needs more than one good photo. It needs a clean main image, a shot of the product at home, a close-up that proves the material, something that shows the size, and a banner when a sale starts.

GPT Image 2.5 came out on September 8. A week later I wrote ten GPT Image 2.5 product photo prompts, one for each of those jobs plus five others, and ran every prompt on both of its models, Flare and Sunburst. Then I checked what a shopper or a store's image rules would catch: spelling on the label, whether the white background is really white, how much of the frame the product fills, and whether a gift set still matches the main photo.

Every product below is invented, and so is the coffee brand on the pouch. If you sell a real product, run the same prompts with your own photo as a reference; that workflow is near the end.

The GPT Image 2.5 product photo formula I used

All ten prompts use the same seven labeled lines in the same order. The labels are for you more than for the model: when a result misses, you know which line to change.

Shot:    what the image is for
Product: what it is, material, color, finish, size
Setting: background, surface and props, with a count
Light:   direction, softness, shadow
Camera:  angle, placement, margins, aspect ratio
Text:    the exact words in quotes, or "None."
Avoid:   what the model tends to add on its own

In product photos, three of those lines need the most care:

• Product gets the most detail. Material, finish and size are what a buyer checks against the parcel.
• Camera describes the framing, but the aspect ratio setting in the generator decides the shape of the file. Set the same ratio in both places, or the crop in your prompt and the canvas won't match.
• Text gets the exact words in quotes. In every run for this post, both models printed quoted text exactly as written, in English and in Chinese.

GPT Image 2.5 main image settings and specs

Store rules differ, so check your own. These are the settings I start from, with what I measured on the files:

Two of those numbers need explaining.

Pure white isn't 255. I checked the edges of all 39 white-background images I generated for this post, from both models, and found no pure 255 white. Most edges averaged about 251, which looks white on a screen. If your store checks for true white, fill the background with 255 white in an image editor before you upload.

Percentages get ignored. I asked for a bottle "filling about 85% of the frame height" and got 93% on Flare and 97% on Sunburst, with the dropper nearly touching the top edge. Asking for 70% only moved it to 89% and 94%. What worked was describing the empty space: "wide empty white margins above and below it, the top of the dropper well below the top edge." In four runs with that wording, across both models, the bottle filled 71–77% of the height. Prompt 1 uses it.

Keep the main image itself free of text, badges and extra props unless your store allows them. Selling points belong on the secondary images, which is what prompts 5 and 6 are for.

Both GPT Image 2.5 models run in the browser on GPT Image 2.5 without a ChatGPT account. An image costs 8 credits at the default medium quality and 1K. High quality, 2K and 4K cost more, and the generator shows the price before you run. New accounts start with free credits. If you need a cutout, generate on white and remove the background afterward; OpenAI's API can also return a transparent PNG or WebP if you call it directly.

10 GPT Image 2.5 product photo prompts

Unless a section says otherwise, each image was generated once per model at medium quality and 1K, and shown as it came out. Copy a prompt, swap the Product line for your item, and keep the rest for the first run.

1. White-background main image

The main image has one job: show exactly what the buyer gets. One product, no props, a soft shadow and plenty of white around it.

Shot: Main product image for an online store listing.
Product: One unbranded 30 ml frosted amber glass serum bottle with a matte black dropper cap and a plain cream label with no writing.
Setting: Pure white seamless background, no props.
Light: Large soft light from the upper left, a gentle contact shadow directly under the bottle, one clean vertical highlight on the glass.
Camera: Front view at eye level, bottle upright and centered with wide empty white margins above and below it, the top of the dropper well below the top edge, square 1:1 crop, edges sharp.
Text: None.
Avoid: Logos, writing on the label, extra bottles, droplets, floating objects, colored backdrop, dark vignette.

Settings: 1:1, 2K, high quality, both models.

Both models centered the bottle on a clean white square with a soft contact shadow and a blank label, and neither added a prop or a droplet. Sunburst's glass looks frosted, as the prompt asks. Flare's is glossier, with brighter reflections along the sides. Either would work as a main image.

2. Lifestyle image

A lifestyle shot shows the product at home, which the main image can't. Name the room, the furniture and the light, and put a number on the props so they don't take over.

Shot: Lifestyle image that shows the product in a real home.
Product: One unbranded handmade ceramic table lamp, matte sage-green glaze with visible throwing rings, rounded base, natural linen drum shade, switched on.
Setting: Walnut side table next to a cream boucle armchair in a calm living room, a folded wool throw and two stacked books with plain covers on the table, a potted plant out of focus behind.
Light: Warm late-afternoon window light from the right, mixed with the lamp's own soft glow.
Camera: Three-quarter view at seated eye level, lamp on the left third, room gently blurred behind it, vertical 4:5 crop.
Text: None.
Avoid: Logos, readable book titles, people, clutter, a second lamp, harsh flash light.

Settings: 4:5, 1K, medium, Sunburst.

The glaze shows its throwing rings, and the window light falls across the floor between the table and the chair. The prompt asks for plain book covers, and neither model printed a title. Flare's version of this prompt was just as usable, with a glossier glaze and flatter light.

3. Flat lay

Flat lays suit clothing, kits and anything sold with several small parts. Two instructions matter most: a camera pointing straight down, and an exact number of props.

Shot: Flat lay for a clothing listing.
Product: One unbranded long-sleeve linen shirt in oatmeal, neatly folded so the collar and top two buttons show.
Setting: Light oak boards. Exactly four accessories around the shirt with even gaps: a woven straw hat, a tan leather belt loosely coiled, a wooden bead bracelet, a small natural canvas pouch.
Light: Soft overcast daylight from above, very light shadows.
Camera: Straight top-down view, shirt in the center, accessories on a loose grid, square 1:1 crop.
Text: None.
Avoid: Logos, brand tags, labels with writing, extra items, wrinkled or dirty fabric, tilted perspective.

Settings: 1:1, 1K, medium, Sunburst.

I asked for exactly four accessories and got four on both models, at a true top-down angle. The two results were close enough that I wouldn't wait for Sunburst on a flat lay. The straw hat runs off the left edge in both. If that bothers you, try adding "every item fully inside the frame" to the Camera line.

4. Detail close-up

A detail shot proves the quality a thumbnail can't. Low light from one side is what makes grain and stitching visible, so say where the light comes from.

Shot: Detail close-up that proves material quality.
Product: One unbranded full-grain leather card wallet in cognac brown with ivory saddle stitching and smooth burnished edges.
Setting: The wallet rests on dark slate; one corner of the wallet fills most of the frame.
Light: Low raking light from the left to show the grain, the depth of each stitch and the polished edge.
Camera: Macro close-up on the corner, stitches sharp in the center, focus falling off toward the back, square 1:1 crop.
Text: None.
Avoid: Logos, embossed stamps, words, cards sticking out, dust, plastic-looking shine.

Settings: 1:1, 1K, medium, Sunburst.

OpenAI lists richer textures among the 2.5 improvements, and leather shows texture well. Both models gave me visible grain, raised stitches and a polished edge. This is also the image type where an invented product causes trouble: the stitch spacing and edge finish here belong to no real wallet. For something you sell, photograph the detail or use your own photo as a reference.

5. Selling-point infographic

Keep each callout to two or three words, and describe the layout in plain terms: product on one side, labels on the other, a line from each label to its part.

Shot: Selling-point image with three short callouts.
Product: One unbranded brushed stainless steel French press, 1 liter, double-wall body, black handle, plunger pressed halfway down.
Setting: Plain warm light-gray background. Product on the left half; three callout labels stacked on the right half, each joined to the product by a thin charcoal line.
Light: Soft studio light from the left, subtle shadow under the product.
Camera: Front three-quarter view, product large but with clear space above and below it, square 1:1 crop, wide clean margins.
Text: Headline at the top, exactly "BREW BETTER AT HOME". Callouts exactly "DOUBLE-WALL STEEL", "4-LAYER FILTER", "1 L CAPACITY". Bold clean sans-serif in charcoal. No other words.
Avoid: Logos, extra text, icons with letters, prices, star ratings, badges, misspellings.

Settings: 1:1, 1K, medium, both models.

Both models spelled the headline and all three callouts correctly and ran each line to the part it names. The product is where they went off. A double-wall steel press is opaque, but both models wanted the filter to show: Flare gave the body a glass window, and Sunburst cut an open slot into the steel. When I ran the Chinese version of this prompt, Flare added the window again and Sunburst kept the body solid. On my next run I'd add "windows or cutaways" to the Avoid line, and I'd still check the shape of the product against the real one.

6. Size reference

Size questions and size complaints cost sellers time, so show scale twice: something familiar inside or next to the product, plus dimension lines.

Shot: Size reference image.
Product: One unbranded natural seagrass storage basket with a cream cotton lining and two rope handles, 40 cm wide and 30 cm tall.
Setting: Light oak floor against a white wall. Three folded white bath towels stacked inside the basket so the size is easy to judge.
Light: Soft daylight from the left, gentle shadow on the floor.
Camera: Straight-on front view at basket height, basket centered, thin charcoal dimension lines with small arrows along the bottom and the right side, square 1:1 crop.
Text: Only two dimension labels, exactly "40 cm" under the bottom line and "30 cm" beside the side line, small clean sans-serif.
Avoid: Logos, extra numbers, rulers, measuring tape, people, tilted camera.

Settings: 1:1, 1K, medium, Flare.

Both models printed "40 cm" and "30 cm" exactly. Neither drew the lines where a tape measure would: in this Flare image both lines stop a little short of the basket's edges, and in Sunburst's image the height line ends well above the base. Use the arrows as a layout guide and redraw them on the final image with your real measurements.

7. Seasonal promo banner

A banner has to leave room for the words. Reserve the empty half in the Setting line, then give the headline and the offer, word for word, in Text.

Shot: Seasonal promotion banner for a store homepage.
Product: Three unbranded handmade stoneware mugs in autumn glazes: rust orange, deep olive and speckled cream, one filled with hot cider.
Setting: Warm terracotta background. Mugs grouped on the right half on a wooden board with a few dried maple leaves and two cinnamon sticks; the left half stays clear for text.
Light: Warm directional light from the right, soft shadows, a thin wisp of steam above the filled mug.
Camera: Slightly elevated front view, wide 16:9 composition, mugs in sharp focus.
Text: Large headline on the left, exactly "THE AUTUMN RESTOCK". Below it, smaller, exactly "20% OFF ALL MUGS · ENDS SEPT 30". Cream serif headline, clean sans-serif subline.
Avoid: Logos, extra words, price tags, fake badges, pumpkins, text over the mugs.

Settings: 16:9, 1K, medium, Sunburst.

The headline, the discount, the middle dot and the date all came out as written on both models, in the type styles I asked for. Swap in your own offer and keep that line short. At 16:9 and 1K the file is 1360×768, small for a full-width homepage banner. Render the final at 2K.

8. Packaging front

For packaged goods, the label is part of the product. Give each line of label text separately and in order, and leave barcodes and certification marks out unless you're supplying real ones.

Shot: Packaging front view for a product page.
Product: One kraft paper stand-up pouch of whole coffee beans with a matte cream front label, invented brand "FERNHOLLOW COFFEE".
Setting: Pure white seamless background, pouch standing upright, eight roasted coffee beans scattered in front of it.
Light: Soft even light from the front left, gentle contact shadow.
Camera: Straight-on front view, pouch centered with empty white space above it and around the beans, square 1:1 crop, label fully readable.
Text: On the label only, exactly three centered lines: "FERNHOLLOW COFFEE", "MEDIUM ROAST · WHOLE BEAN", "250 g". Simple dark-brown sans-serif.
Avoid: Other words, logos, barcodes, certification seals, nutrition panels, a tilted pouch.

Settings: 1:1, 1K, medium, both models.

All three label lines came out exactly as written on both models, middle dot included. The rest of the prompt was hit and miss. Flare left white space above the pouch; Sunburst's pouch still fills most of the frame height. The prompt also asks for eight coffee beans. In this pair Flare gave me eight and Sunburst seven; in the Chinese run it was the other way around. Across the runs of this prompt both models came up a bean short at least once, so count the props on every image, not only the first one.

9. Gift set, built from the main image

A set shot is where a listing's photos usually stop matching. Instead of describing the bottle again, upload your approved main image in the image-to-image generator and point to it in the Product line.

Shot: Gift set image that shows everything in the box.
Product: The serum bottle from the uploaded photo, kept identical in shape, glass color, cap and label, plus three matching pieces in the same frosted amber glass with matte black caps and plain cream labels: a 50 ml cream jar, a 100 ml pump bottle and a 10 ml roll-on bottle.
Setting: An open kraft gift box with a cream paper insert holding all four pieces upright in one row, the lid leaning behind the box, pure white background.
Light: Soft light from the upper left, the same shadow direction on every piece.
Camera: Front view from slightly above, all four pieces fully visible, square 1:1 crop.
Text: None.
Avoid: Logos, writing on labels, mismatched glass colors, extra products, ribbons covering the pieces.

Settings: 1:1, 1K, medium, Sunburst, with the Sunburst image from an earlier run of prompt 1 uploaded as the reference.

The left set came from the same prompt with the serum bottle described in words instead of uploaded. Its bottle is darker, the label is shorter and the shoulders are narrower than in the main image. With the main image uploaded, the bottle kept its shape, glass color and tall label. The finish still drifted a little from frosted toward glossy, so compare the finish as well as the shape.

10. Hands-only shot

Hands show size and use while keeping attention on the product. Say that only hands and wrists are in frame, then rule out the details that usually go wrong.

Shot: Hands-only image that shows real size and how the product is used.
Product: One unbranded small square jewelry box in sage-green velvet, lid open, holding a pair of thin gold hoop earrings.
Setting: Neutral linen fabric background, soft and out of focus.
Light: Soft window light from the left, warm highlights on the gold.
Camera: Close crop from slightly above; two relaxed hands with short natural nails hold the open box toward the camera; only hands and wrists in frame; box in sharp focus; vertical 4:5 crop.
Text: None.
Avoid: Faces, rings on the fingers, nail art, extra fingers, logos, writing inside the lid.

Settings: 4:5, 1K, medium, Sunburst.

Both models gave me relaxed, natural hands with no extra fingers and no rings, and the earrings sit properly in the slot. Check the fingers anyway. In any image with hands, that's still the first place to look.

Which GPT Image 2.5 model I use for product photos

The two models were close on product photos, and they cost the same credits here, so the choice mostly comes down to how long you want to wait. At 1K and medium quality, Flare returned each of the nine first-round images in 17 to 28 seconds, and Sunburst took 30 to 38. At 2K and high quality, prompt 1 took 41 seconds on Flare and 71 on Sunburst.

I draft on Flare: prompt changes, scene ideas, most secondary images. I move to Sunburst for the lead image once the prompt is settled, and for shots that sell the material. On the lamp and the serum bottle, Sunburst kept the matte and frosted finishes I asked for, and Flare's came out glossier. It isn't a clean win, though. Sunburst also cut a slot into the French press and left almost no space above the coffee pouch, so its images get the same checks.

For tests beyond product photos, see GPT Image 2.5 Flare vs Sunburst: 6 Same-Prompt Tests.

What GPT Image 2.5 still gets wrong in product photos

Text isn't on the list below. Beyond the ten prompts, I tried two denser labels at 1K: a twelve-ingredient list on the back of the serum bottle, and an eight-line back label on the coffee pouch in Chinese. Both models got every character right. The model prints whatever you type, though, so copy ingredient lists, dates and claims from your real packaging.

The mistakes were in the product, the layout and the background:

• Construction. When a callout names a part you can't see, the model may change the product to show it (prompt 5).
• Counts. Eight coffee beans came out as seven in several runs, on both models (prompt 8).
• Measurements. Dimension labels read correctly, but the lines don't match the object's edges (prompt 6).
• Finish. A reference keeps the shape more reliably than the surface finish (prompt 9).
• Angle. An uploaded photo tends to keep its own camera angle (the label test below).
• Background. "Pure white" comes out near-white, not 255 (the settings table).

GPT Image 2.5 reference photo workflow for a full listing

If you're selling a real product, the photos have to match what arrives in the box, so start from your own photo.

1. Take one clear photo of the real product. Plain background, even light, label facing the camera. It doesn't need to be a studio shot; it needs to show the shape, color and label clearly.
2. Make the main image from it. Open the image-to-image generator, choose a GPT Image 2.5 model, upload the photo, and run prompt 1 with the Product line changed to "The product from the uploaded photo, kept identical in shape, color and label." Set 1:1 and 2K.
3. Use the approved main image for everything else. Upload it as the reference for the lifestyle, set and hands prompts. The generator takes up to 16 reference images and makes up to 8 images per run, so one run gives you several versions of a scene to choose from. Proofread the label on every result.

OpenAI says 2.5 keeps subjects from reference photos better than Images 2.0, so I tested labels. I uploaded an earlier Sunburst run of prompt 8 and asked for a kitchen scene from a three-quarter angle. Both models kept all three label lines word for word. Neither gave me a real three-quarter view: the pouch turned only slightly, with the label still facing the camera. The Chinese version of the test went the same way. For a product page that's a fair trade, but don't count on a reference photo to give you a new camera angle.

The Bottom Line

A product photo prompt works best when it reads like a shot list: what the image is for, what the product is made of, the setting, the light, the framing, the exact words and what to leave out.

In my runs, GPT Image 2.5 got the words right every time. The product needed more checking than the text did. Start from a real photo when the product is real, describe margins instead of percentages, and compare every image with the thing you're shipping.

Related reading

• GPT Image 2.5 Flare vs Sunburst: 6 Same-Prompt Tests
• 8 GPT Image 2 Design Prompts for Real-World Work
• 10 GPT Image 2 Poster Prompts I Use for Client-Ready Designs

gpt-image2.art is an independent site and is not affiliated with OpenAI. The products and the coffee brand in this post are invented.