By René Mattis · vona ·
How this article is made
Topic, structure, practical experience and figures come from René Mattis. For wording, researching sources and the English translation, he works with an AI (Claude by Anthropic). He checks every statement, every figure and every source before the article is published. We disclose this because we stand for a transparent use of AI – and because this is exactly the way of working we also implement for our clients.
AI Product Images: Why the Subscription Cost More Than Our Own Solution
Case study LANAKILA: how a platform subscription for AI product images became a custom desktop app connected directly to Google's image AI – and what that means per image.
In the article Reducing ongoing AI costs we looked at the principle: costs rarely come from the model itself, but from the way it is used. Here we show the same thing with a concrete case from our work: AI product images for an online shop. It is about LANAKILA, the sportswear brand of our owner Sylvia Michalk. So this is not an independent client, but a real case with real figures. For every figure we say where it comes from.
The starting point
LANAKILA designs triathlon suits, cycling jackets and other sportswear. New collections were to be offered in the shop before the photo shoot had taken place. That requires product images that show the new design realistically, ideally on a model and in the style of the existing photos.
First attempt: a platform subscription
After comparing several providers, LANAKILA chose a well-known online platform for AI images and videos. We deliberately do not name it, because the pattern is similar on many platforms. LANAKILA took out a Pro subscription. According to the platform’s pricing page it costs USD 25 a month and includes 600 credits, advertised as “up to 300 images”.
In practice, things looked different:
- Resolution costs extra: For product images LANAKILA needed a higher resolution (2K). Instead of the advertised up to 300, LANAKILA recalls getting only around 30 images a month. How many credits an image costs at which resolution is not stated on the pricing page. The “up to” evidently applies to the cheapest setting.
- Corrections cost too: It often took several attempts before a product image looked the way it should, and every attempt uses credits.
Mathematically, a 2K image thus cost around USD 0.83, and a usable one even more because of the corrections. After a little over a month the subscription was cancelled.
The analysis: what is behind the images?
Platforms like this are usually built on image models from large providers. The platform bundles the model, interface and storage and prices all of it into its credits. So we looked at what it costs to use the models directly. Google’s Gemini image models can be accessed through an API, set up via Google AI Studio. Billing is per generated image, with no subscription and no minimum. Google’s list prices (as of October 2026):
| Model | 1K | 2K | 2K via batch |
|---|---|---|---|
| Nano Banana Pro (Gemini 3 Pro Image) | USD 0.134 | USD 0.134 | USD 0.067 |
| Nano Banana 2 (Gemini 3.1 Flash Image) | USD 0.067 | USD 0.101 | USD 0.050 |
Small amounts for the input – instructions and reference images – come on top. Via the batch interface an image costs half, if the result does not have to be ready immediately, for example for a whole series overnight.
Before recommending anything, we tested the API with real LANAKILA motifs: how well does the model apply a new design to an existing product photo, how many attempts does it take, and what does it actually cost? The result was convincing: the quality was right, and the cost per image was a fraction.
The decision: no new platform, but a tool of our own
The obvious step would have been to simply choose another platform. Instead we recommended a custom solution, because LANAKILA needs images regularly and several people work with them. The result was RENEKI, a desktop app for macOS and Windows tailored exactly to this workflow:
- Working with your own templates: the new design comes straight from the graphics program as the main image, an existing product photo serves as the reference. The AI “dresses” the model in the new design.
- Prompt building blocks at a click for camera, light, style and background. This reduces failed attempts because proven instructions are reused.
- Straight into the shop: once an image fits, it is assigned to a product in the shop, or an existing item is cloned with the new images.
- As a team: projects can be shared between several installations, users and models are managed centrally.
- Data stays on the computer: designs and pattern files are not uploaded to a third-party platform; only the images needed for each generation go to Google.
The result in numbers
| Platform subscription | Own app with the Gemini API | |
|---|---|---|
| Cost per 2K image | about USD 0.83 (USD 25 for about 30 images, according to LANAKILA) | USD 0.134 with Nano Banana Pro (list price), USD 0.067 via batch |
| 30 images in 2K | USD 25 | about USD 4 |
| Billing | monthly subscription, credits expire after one year | only what is generated, no base fee |
A 2K image therefore costs about one sixth, via batch about one twelfth. And the actual costs? According to Google’s billing, LANAKILA paid a total of EUR 38.34 for the image AI in the 28 days from 13 September to 10 October 2026, for two workstations. On one of the two computers alone, 152 images were created in that period, most of them with Nano Banana Pro in 2K (counted in RENEKI; deleted attempts are not included). With the previous subscription, that many 2K images would have required about five Pro subscriptions, roughly USD 125 a month, not counting the second workstation.
For cost control, a monthly spending cap is set at Google. It protects against outliers; according to Google, it may take effect with a delay of around ten minutes.
To be fair, this belongs here too: developing your own app costs time and money, and it needs maintaining. For a handful of images per quarter it is not worth it; a simple interface for the API or a platform with a suitable plan is often enough. For LANAKILA it pays off because new designs keep coming, several people work with it and the shop connection saves manual work every time.
What to watch out for with AI product images
- The image must show the real product. Colours, cut and details have to be right. An AI image that shows a product nicer than it is is not only annoying for customers but can also be misleading under competition law. Every publication should therefore be preceded by a careful check; RENEKI has a team approval step for this.
- Check the labelling. The transparency obligations of the EU AI Act have applied since 2 August 2026. Anyone publishing realistic AI images that look like genuine recordings of existing people, places or objects may have to label them as artificially generated. Whether this applies to AI product images in a shop depends on the individual case; when in doubt, a note such as “AI-generated product image” is the safe choice. More on our EU AI Act page.
Lessons learned
- Read “up to” carefully: credit models often assume the cheapest setting. What matters is the price per usable result at the quality you really need.
- Test yourself instead of comparing promises: a short test with real motifs shows quality, attempts and cost per image.
- Go to the model, not just the interface: if you use AI regularly, check what the model provider’s own API costs.
- Build the tool around the workflow: a tool that fits your own workflow saves not only fees but also corrections and manual work.
- Cap and measure costs: set a spending cap at the provider and keep an eye on the cost per image.
If you are running AI tools or platform subscriptions and want to know whether it could be cheaper and better, let’s look at it together. Under AI consulting you will find our review of existing AI processes and costs, under app development custom tools like RENEKI. Or go straight to a free initial consultation.
More about LANAKILA and its collections: lanakilasports.de (opens in a new tab)
As of October 2026. Platform price according to its pricing page, retrieved on 10 October 2026; number of 2K images in the subscription according to LANAKILA (from memory, billing is no longer accessible after cancellation). Google prices according to the Gemini API pricing page (as of 9 October 2026), spending according to LANAKILA’s Google AI Studio billing, image count from RENEKI.