Yes, to a large extent. AI is strong at backgrounds, lifestyle images, seasonal variants and formats built around a real photo of your product. The packshot itself is still best made with a camera: if you have the product generated, sooner or later the label, the shape or the colour will be off.
In our studio, alongside AI video production, we also create AI visuals: we generate and edit images with Nano Banana, and set the look in advance with Midjourney [1].
What AI does well with product photos, and what it doesn't
- Good, with checks: cutting out products and replacing or extending backgrounds. Watch the edges of glass and hair, shadow and reflection.
- Good for ads, social and extra shop images: lifestyle images and seasonal variants. Watch the scale, and details that shift from one variant to the next.
- Be careful: your product in a hand or with a model. Fingers and grip remain a weak point, and a realistic AI person may require a label [2][3]. Google's free image tool Product Studio doesn't even support images with hands [4].
- Don't: have the product itself generated or the text on the packaging changed. The label, shape, colour and lettering will then deviate; better to adjust the real label and photograph it again.
What does an AI product photo cost compared with a photographer?
| Option | Price | Note |
|---|---|---|
| Nano Banana 2 (Gemini API) | $0.067 (1K), $0.101 (2K) per image | Per attempt [5] |
| Nano Banana Pro (Gemini API) | $0.134 (1K or 2K), $0.24 (4K) per image | Per attempt [5] |
| Productfotografie.com (NL) | €4.50 to €9.50 per photo on a white background, depending on quantity | VAT status not stated [6] |
| Productfoto.nl (NL) | €17 per packshot, €75 set-up fee per project | VAT status not stated [7] |
| Fotofy (NL) | €29 for the first photo per product, €9 per extra photo | Excl. VAT and project costs [8] |
We found no Belgian product photographers with public prices; the Dutch ones give an order of magnitude. An example calculation: 40 products with three lifestyle images each and five attempts per image (an assumption, not a measurement) cost $80.40 in compute with Nano Banana Pro [5]. The 40 packshots cost €755 at Productfoto.nl [7].
The hours spent choosing and checking each image weigh more heavily than the compute. We work out the same cost model for video in what does a video ad cost?
How to make an AI product photo
For backgrounds and scenes around a real photo, tools such as Nano Banana, Photoroom and Adobe Firefly work well. Watch the licence: you may not use a free Photoroom account commercially [9], and Adobe only calls its own Firefly models commercially safe; you assess partner models yourself [10].
- Start with a real, sharp packshot. For Shopping, Google recommends at least 1500 × 1500 pixels and asks you not to upscale images [11]. Photograph each colour or variant separately.
- Cut out the product and check the edges at 100%.
- Describe the scene with your packshot as a reference: subject, setting, light, angle and what stays fixed ("keep the shape, colour and logo exactly as in the reference"). Make the light in the scene match the light on your packshot.
- Check every image for label and logo, colour, proportions, shadow, material and details. If the scene suggests a function or quality the product doesn't have, drop the image.
- Export per channel, and leave the AI metadata in the file.
The same check, applied to video, is in from product photo to video ad with AI.
Which rules apply per channel?
| Channel | What applies to the image | Specific to AI |
|---|---|---|
| Google Shopping | The complete, correct product with little or no styling; no overlays or borders; maximum 16 MB [11] | Metadata stating that the image was created with generative AI, such as IPTC DigitalSourceType TrainedAlgorithmicMedia; don't remove it [11] |
| bol.com | Main image on a white background without shadow, models, text or logos; since 1 July 2025, bol.com only enforces this for self-promotion [12] | No separate AI rule [12] |
| Meta ads | The usual advertising rules | "AI info" label on ads made with Meta's own AI tools; Meta has announced detection of other tools [3] |
| Your own online shop | No platform rule | Consumer law: misleading practices are prohibited [13] |
Google's metadata requirement is easy to miss. An image you edit in another program after generating it and then export again can lose its metadata, so check that before you update your feed. How to build that feed and your campaign is explained by Benjamin in setting up a Google Shopping campaign.
Does an AI product photo need a label?
As a rule, no, as long as the image doesn't mislead about your product. The European Commission cites colour correction, extending or replacing a background and resizing images for an advertisement as examples of what is not a deepfake [14]. An AI image in which your product looks different, more attractive or better than in reality is one [14].
The Raad voor Reclame, the Belgian advertising council, draws the same line, and adds that an exaggerated product image remains misleading, even with a label [2]. The Belgian Advertising Code sums it up: the mere use of AI does not in itself require a disclosure [15]. The label texts in Dutch and French are in Does your video ad need an AI label?
From product photo to video
Good packshots and lifestyle images are also the building blocks of a video ad: each image can become a shot, with camera movement, sound and a voice. In the studio, we call those images statics.
Want to see what your product photos turn into as video? Send your statics for a free test ad: one video ad on your own brand, at no cost or obligation.
Frequently asked questions
Can I create product photos entirely with AI?
You can, but we advise against it. Start from a real photo and use AI for the background, scene and variants. If you have the product itself generated, the label, shape or colour easily deviates, and an embellished product image is misleading.
What does an AI product photo cost?
The compute is small: via Google's API, an image with Nano Banana 2 costs $0.067 in 1K, and with Nano Banana Pro $0.134. Published photographer prices range from €4.50 to €29 per photo. The real cost of AI lies in the checking.
Can I use AI product photos in Google Shopping?
Yes, if each AI image contains metadata stating that it was created with generative AI, such as the IPTC tag DigitalSourceType TrainedAlgorithmicMedia, and you don't remove it. The usual requirements also apply: the complete, correct product, without overlays or borders.
Does an AI product photo need a label?
A real product on an AI background or in a different format generally needs no label, as long as the image doesn't mislead. An image that shows your product as nicer or better than it is counts as a deepfake for the European Commission.
Can I use an AI lifestyle image as the main photo on bol.com?
Better not. For the main image, bol.com requires a white, neutral background without shadow, models, text or logos, and no lifestyle image. Lifestyle images belong among the additional images.
Sources
- AI video production (studio page: Nano Banana and Midjourney in the stack, test ad) — Izoard, accessed 27 Sep 2026
- AI in advertising: transparency, disclosure and 12 clear principles (adapting packshots to formats, product performance) — Raad voor Reclame, 2026-08, accessed 27 Sep 2026
- Expanding GenAI Transparency for Meta's Ads Products — Meta, 2025-02-03, updated 2026-06-01, accessed 27 Sep 2026
- About Product Studio — Google Merchant Center Help, accessed 27 Sep 2026
- Gemini Developer API pricing (Nano Banana 2, Nano Banana Pro) — Google AI for Developers, accessed 27 Sep 2026
- Prijzen — Productfotografie.com, accessed 27 Sep 2026
- Productfotografie prijzen — Productfoto.nl, accessed 27 Sep 2026
- Tarieven — Fotofy, accessed 27 Sep 2026
- Free accounts and commercial use — Photoroom Help Center, 13 Oct 2025, accessed 27 Sep 2026
- Adobe's approach to generative AI (Firefly) — Adobe, accessed 27 Sep 2026
- Image link [image_link] — Google Merchant Center Help, accessed 27 Sep 2026
- Deze regels gelden voor je afbeeldingen — bol.com Partnerplatform, accessed 27 Sep 2026
- Reclame en oneerlijke praktijken ten aanzien van de consumenten — FPS Economy, 29 Sep 2025, accessed 27 Sep 2026
- Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, C(2026) 5054 (point 116 and the examples under 6.1.1) — European Commission, 20 Jul 2026, accessed 27 Sep 2026
- De Belgische Reclamecode (Belgian Advertising Code), version 22 (Art. 9.1: transparency) — Raad voor Reclame, 2026-09, accessed 27 Sep 2026