Use real photography wherever the image must show exactly what the customer will receive, and consider AI-generated imagery for backgrounds, settings and concept visuals around a real product. The deciding factors are accuracy, the rules of the place where the image will appear, and consumer law. This guide compares the two approaches for stills and ends with a decision framework and a hybrid workflow.
Start with the question that matters: what must the image prove?
A product image does two jobs. It attracts attention, and it tells the buyer what they will get. Photography is strong at the second job because a camera records the actual object. Generative tools are strong at the first job because they can produce settings and moods that would be expensive to build physically. The risk with AI imagery is not that it looks poor; it is that it can look excellent while showing something subtly different from the product.
Break your image needs into two groups:
- Evidence images: main listing image, detail views, colour and finish, scale, packaging, what is in the box.
- Atmosphere images: lifestyle scenes, hero banners, social content, campaign visuals, seasonal backgrounds.
Evidence images should come from the real product. Atmosphere images have more room for generation, provided the product within them stays true.
Accuracy and Australian Consumer Law
The ACCC states that business claims must be accurate, including images and descriptions of what is offered, that businesses should check the overall general impression is accurate, and that it makes no difference whether a business intends to mislead (ACCC). Applied to product images, the points to check include:
- Does the colour match the real product under normal viewing?
- Are texture, finish, size and proportions faithful?
- Is every item shown actually included in the sale?
- Are features, ports, stitching, labels or fittings rendered correctly, not invented?
- Has retouching or generation made the product look better than it is?
Generated images can add features, change a logo or smooth a surface without anyone noticing. Someone must compare the final image against the physical product. This is general information, not legal advice; seek professional advice if you are unsure about a particular claim.
Marketplace and shopping-feed rules
Platforms set their own image rules, and they change. At the time of writing (November 2026), Google Merchant Center’s image requirements include the following (Google Merchant Center Help):
- The image must offer an unobstructed, clean view of the item for sale, and must show the exact item for sale.
- Promotional elements such as calls to action, prices, watermarks and superimposed logos are not allowed. Brand marks are allowed only if they are a physical part of the product.
- Placeholder graphics, shared stand-in images and illustrations or mockups that do not depict the physical product are not allowed (with limited category exceptions listed on that page).
- Products not sold with the main item must not appear in the image.
- The minimum is 500 x 500 pixels, which the page says becomes required for all products from 31 January 2027; about 1500 x 1500 pixels or larger is recommended.
- Images created with generative AI must keep embedded metadata identifying them as AI-generated, such as the IPTC DigitalSourceType tag.
Google’s separate AI-generated content page adds that this metadata should be kept, not stripped, even when images come from tools such as Product Studio, and that AI-generated titles and descriptions use structured attributes (Google Merchant Center Help).
Two implications follow. First, a purely synthetic image of a product is difficult to justify as a main image where the rule is that it shows the exact item. Second, if you use any generative step, the export pipeline must keep the metadata. Many editing and compression tools strip it, so test the final file. Other marketplaces (including Amazon) publish their own rules in their seller help centres, which sit behind seller log-ins; read the current version for each category you sell in before commissioning images.
Comparing the approaches
| Factor | Real photography | AI-generated imagery |
|---|---|---|
| Accuracy of the product | High, it records the real item | Variable, needs checking against the product |
| Marketplace main images | Generally the safe route | Risky where the exact item must be shown |
| Settings and mood | Costs rise with sets, props, locations | Flexible, with consistency work |
| Consistency across a range | Controlled by a repeatable setup | Needs references and review |
| Revisions | New shoot or retouch | Fast iterations, but each needs checks |
| Disclosure and metadata | None required for a real photo | Metadata handling and platform labelling |
| Rights questions | Straightforward for your own shoot | Check the tool’s terms and the status of the output |
For deeper discussion of rights and legal limits of generated output, see the real limitations of AI-generated video in commercial production, which also applies to stills.
Cost drivers (without numbers)
Neither approach has a universal price advantage. The drivers differ.
Photography cost drivers: number of products and variants, styling and props, location or set build, model or hand talent, retouching depth, usage rights and the need to reshoot when products change.
AI imagery cost drivers: the effort of getting a faithful product into the scene (reference photos, masking, compositing), consistency across a range, the number of review rounds, correction time for defects, and specialist skills to prepare compliant files.
A catalogue of hundreds of near-identical items is often cheaper to photograph on a repeatable setup. A small number of high-concept campaign visuals, where a physical set would be impractical, may suit generation around a real product.
The hybrid approach: real product, generated surroundings
The workflow that tends to balance accuracy and flexibility starts with the real photograph.
- Photograph the product properly: consistent lighting, multiple angles, accurate colour. These are your reference and your evidence images.
- Cut out and retain the real product pixels. The product itself is not regenerated.
- Generate or build the background and setting, guided by a brand code sheet and mood reference.
- Match light and shadow so the product sits believably in the scene.
- Review against the physical product for colour, logo, proportion and details.
- Check platform fit: which images are allowed as main images, and what labels or metadata apply.
- Preserve metadata and record provenance (which images are real, composite or generated).
For moving images built from the same principle, see AI product videos: from concept to campaign.
Decision framework
Ask in order:
- Is this an evidence image or an atmosphere image? Evidence: photograph.
- Does the destination require the exact item, with strict rules? Photograph, or a composite that keeps the real product.
- Would a physical set be impractical or wasteful for this concept? Consider generation around a real product.
- Can you verify the final against the physical item before publishing? If not, do not use generated product detail.
- Is the brand look consistent with existing assets? See what makes a brand look premium for what to check.
Where Radiance Impact fits
Radiance Impact is a Brisbane-based creative technology studio. Its Studio division produces AI commercials and product films, brand films and trailers, short-form and social video, and post-production. If you are weighing a real shoot against generated or composite visuals for a product launch, contact us to talk through the options.
Sources
- Image requirements for product listings, Google Merchant Center Help
- AI-generated content, Google Merchant Center Help
- False or misleading claims, Australian Competition and Consumer Commission
