Blog
Consistent product photos with AI: one pack shot, many settings
A stoneware mug moved from a studio sweep to a kitchen counter for 8 credits, keeping its glaze, speckle, cobalt band and handle. The Object role, what it holds, and what a catalogue costs in credits.
To get the same product in many AI photographs, make one clean pack shot, then attach it to every later generation with the Object role — the one that tells the model to reproduce its shape, proportions, markings and materials exactly. We did it with a stoneware mug for 8 credits a shot. The glaze, the speckle, the cobalt band and the thumb-rest on the handle all survived the move from a studio sweep to a kitchen counter.
The pack shot, and the lifestyle shot from it
The prompt for the second was one sentence of new direction — oak counter, linen cloth, morning window light from the right, three-quarter from slightly above — sitting under the sentence the Object role writes for you:
Reference 1 is an object — reproduce it exactly: its shape, proportions, markings and materials.
What held, and the one thing that moved
| Feature | Survived? |
|---|---|
| Silhouette: straight-sided body, the handle's curve | Yes |
| The raised thumb-rest at the top of the handle | Yes — the detail most likely to be dropped |
| Matte oatmeal glaze and its fine dark speckle | Yes |
| The single cobalt band near the rim | Yes, but painted a little thicker and less even |
| Proportions | Close; the second shot reads very slightly taller |
For a catalogue that is a pass. For a product where the band's exact weight is the brand, it is a near-miss you would notice at scale — so check the identifying detail on every shot before you commit a set.
Why the Object role and not a style reference
Gemini has no typed reference slots. Every attached image lands in one flat array with the prompt appended last, and nothing tells the model which picture is the product. It infers the job from the words. Attach your pack shot with the Style role and you are explicitly telling it to take the palette and ignore the subject — you will get a mug, but not your mug. That single mislabel is the most common reason product consistency "does not work".
A catalogue, priced
- Make one pack shot on a plain sweep, flat even light, the product filling a good part of the frame. This is your master; reuse it for everything.
- Attach it with the Object role on every later generation.
- Change only the setting clause between shots — counter, shelf, hand, outdoor table. Keep the lighting words alike if the set must look like one shoot.
- Name the identifying detail in words as well: "one cobalt band near the rim, a raised thumb-rest on the handle".
- Check each shot at 1K for 8 credits, then upscale the keepers for 15.
| Set | Generations | Credits at 1K | With the keepers upscaled |
|---|---|---|---|
| Pack shot + 4 settings | 5 | 40 | 40 + 3 × 15 = 85 |
| A 10-SKU line, 3 shots each | 30 | 240 | 240 + 10 × 15 = 390 |
A 10-SKU line at three shots each is 240 credits — inside a single $20 month, with 300 left. The comparison worth making is not against another generator; it is against a day of studio time.
The honest limits
- Text on packaging is a separate problem. If your label has words, generate at Nano Banana Pro and check every character — it is built for legible in-image text, and it is still worth reading twice.
- Exact colour is not guaranteed. Our cobalt shifted slightly in weight. For a brand colour, extract the hex from the master's decode and name it in the prompt.
- This is not a photograph of your product. It is a picture of something that looks like it. Where a marketplace or an advertising standard requires a real photograph of the goods, this is the wrong tool.
- Basic allows 3 reference images per generation, Pro 10. Two or three angles of the same product as references hold the shape much better than one.
FAQ
How do I get consistent product photos with AI?
Shoot one clean pack shot, then attach it as a reference with the Object role on every later generation. That role tells the model to reproduce the shape, proportions, markings and materials exactly, and it is what keeps the product the same product.
Why does my product change between shots?
Usually because the reference is unlabelled, or labelled as a style reference — a role that explicitly tells the model to ignore the subject. Check the badge on the thumbnail before you rewrite the prompt.
What does a product set cost?
8 credits a shot at 1K on Nano Banana Pro. A pack shot plus four settings is 40 credits; a ten-product line at three shots each is 240, which fits inside one $20 month with 300 credits to spare.
Will it keep small details like a logo or a painted stripe?
Mostly. Ours kept the glaze, the speckle and the raised thumb-rest, and painted the cobalt band slightly thicker than the master. Name the identifying detail in words as well as relying on the reference, and check every shot before committing a set.
Can I use these instead of photographing the real product?
For mood, lifestyle and concept work, yes. Where a marketplace or an advertising rule requires an actual photograph of the goods you are selling, no — this is a picture of something that looks like your product, not of your product.
Make one pack shot in the Studio, badge it Object, and move it to one new setting — that round trip is the whole method. The other five reference roles are in part 4 of the generation series.