Overs

How to start using AI for product photos this week

Checked 5 min read

Short answer

Give it five days and one product. Day 1, photograph it; day 2, write one page of brand notes; day 3, make ten renders; day 4, check them against the real product and remake the near misses; day 5, publish one image in one place, next to a real photo. A week of about 30 renders costs under $7 at Google’s and OpenAI’s published prices, and free apps let you try it for nothing.

The five-day plan

DayDo thisYou end with
1Photograph one product: every side, the label straight on, a texture close-up, a size shotA reference set of 8 to 10 photos
2Write brand notes: colors with hex codes, light, camera, props, a never listOne page you paste into every prompt
3Choose one channel and make 10 renders of two or three shot ideas10 renders to judge
4Check each render against the real product; remake the near misses with one note eachA few keepers, or a clear reason there are none
5Publish one keeper in one place, next to a real photoA live test

Photograph one product (day 1)

Pick a best seller with a simple shape and little text on it. A plain bottle or box is easier than a watch or a sheer dress, so the first week tests the method instead of your hardest product. Shoot it in window light, zoomed to 2x, from every side. The full method is in how to photograph your product for AI.

Write one page of brand notes (day 2)

The model knows nothing about your brand until you tell it. Write your look down in words and numbers it can follow, and paste the same page into every prompt.

Brand notes, one page
Brand: [name], [what you sell] for [who]
Palette: [color name] [hex], [color name] [hex], [color name] [hex]
Light: [for example, soft window light from the left, morning]
Camera: [for example, 50 mm at table height, product in the right third]
Surfaces and props: [materials you use], three props at most
Grain and color: [for example, light film grain, natural color, never oversaturated]
Never: [for example, other brands, text in the scene, marble, confetti]

For a fuller version, see how to write brand rules an AI can follow.

Make ten renders (day 3)

  1. Choose one channel: a product page image, an Instagram post or one ad. One place means one size and one set of rules.
  2. Write two or three shot ideas, each as a short scene brief. There are templates in how to make lifestyle product photos with AI.
  3. Draft on a cheap model or a low-quality setting, then remake the best drafts on a better one.
  4. Send only the reference photos each shot needs, with the product photo first.
  5. Save every render, and note which prompt and model made it.

Review and remake (day 4)

  • Label, logo, colors, shape and parts match the real product, held side by side
  • The product has a shadow where it touches, and the light comes from one direction
  • The image follows your brand notes
  • You would post it next to your real photos

Sort each render into keep, fix or drop. Remake a fix with one specific note, such as "move the light to the left" or "the cap is matte black", instead of a new prompt. Then divide renders by keepers. That ratio is your first real cost number; across our agency’s four 2026 campaigns it averaged about 8, as shown in how many AI images it takes to get one you can use. The full review list is the QA checklist for AI images.

Publish one test (day 5)

Put one keeper where you can compare it with a real photo: the second image on a product page, one post, or one ad running beside your current best ad at the same budget. Change one thing at a time, so the result tells you something.

Judge it on two things. First, whether anyone noticed something wrong with the product, in comments, returns or support emails. Second, whether it did its job, in clicks, saves or add-to-carts, as well as the real photo beside it. For a fair ad test, see how to tell if AI creative is working.

What the first week costs

The plan above makes about 30 renders: 10 on day 3, and on day 4 five near misses remade four times each. At the prices Google and OpenAI publish:

ModelPer render10 renders30 renders
GPT Image 2, low quality (OpenAI), for drafts$0.006$0.06$0.18
Nano Banana 2 Lite (Google)$0.0336$0.34$1.01
GPT Image 2, medium quality (OpenAI)$0.053$0.53$1.59
Nano Banana 2 (Google, 1K)$0.067$0.67$2.01
Nano Banana Pro (Google, 1K or 2K)$0.134$1.34$4.02
GPT Image 2, high quality (OpenAI)$0.211$2.11$6.33
Published prices, September 2026. OpenAI prices are for a 1024 x 1024 image. Tool fees and your time are extra.

Google’s API has no free tier for these image models, so API use is billed from the first image. If you buy model credit through OpenRouter, a service that gives you one account and one bill for many AI models, a card purchase adds 5.5% with a $0.80 minimum, so a $10 top-up costs $10.80.

To spend nothing, use a free app. The Gemini app makes images with Nano Banana 2 on every plan, including with no paid plan. OpenAI rolled out ChatGPT Images 2.5 to every ChatGPT plan on September 8, 2026. Google Pomelli, free from Google Labs, turns a product photo into studio or lifestyle shots from templates; it launched in the US, Canada, Australia and New Zealand. Chat apps work one image at a time, and Pomelli works from templates. See free AI product photo tools.

Questions people also ask

Do I need to know how to write prompts?
Not much at first. Write what a photographer would need to know: the product, the surface, where the light comes from, the lens and the colors. Your brand notes page covers most of it, and the scene brief covers the rest.
Which model should I start with?
A cheap one, and switch only when it fails your checks. In September 2026 Google’s Nano Banana 2 Lite cost about 3 cents a render, while Nano Banana Pro and GPT Image 2 at high quality cost 13 to 21 cents. Draft cheap and finish on the better model.
Do I have to tell customers the photos are AI?
That depends on where you sell and advertise. Google Merchant Center requires AI-made product images to keep their AI metadata, and its help center notes that AI rules in the EU, India and New York require disclosures or labels on certain ads with AI-made or AI-edited images. Check the rules for each channel before day 5.

Where Overs fits

Overs runs the same order for you: brand research and rules first, then a shot plan with an estimated cost that you approve before anything is made. The plan is free for 40 photos a month. The AI costs a few cents a photo through your own OpenRouter account, a service that gives you one account and one bill for many AI models, so you need a little credit there before the first photo.

Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.

Sources

Checked on September 24, 2026. Prices, specs and rules change; follow the links for the current versions.

  1. 1.Google AI for Developers: Gemini API pricing
  2. 2.OpenAI API docs: Image generation (cost table)
  3. 3.OpenRouter: FAQ (fees)
  4. 4.Gemini Apps Help: Gemini Apps limits and upgrades for Google AI subscribers
  5. 5.Google: Create on-brand marketing content with Pomelli (October 28, 2025)
  6. 6.Google: Pomelli Photoshoot (February 19, 2026)
  7. 7.9to5Mac: OpenAI releases ChatGPT Images 2.5 (September 8, 2026)
  8. 8.Google Merchant Center Help: AI-generated content
  9. 9.Google Merchant Center Help: Allow Merchant Center to improve your product images (AI labels)