Fall Release: GPT Image 2.5 & Your Own Studio Set

Brandmachine takes a product from its first photo to a finished file that goes live in your shop. Generating the image is one step of that, and most of my time goes into the steps around it.

I still talk to brands every week who tried AI imagery and never got it into production. When I ask why, it comes down to two things. The first is scale. One good image is easy, but a catalog means thousands of them, planned, checked, named and delivered the same way every time. The second is quality, and that's what most of this release is about.

OpenAI Image 2.5 Release - Text and logos, at a price that works

If your products carry a logo, a print or text on a label, the Google models have not been good enough so far. OpenAI's model was, but only on its high tier, at €0.70 an image. A finished image usually takes several renders, so on a real catalog that price added up fast.

GPT Image 2.5 changes that. It now renders every GPT image in Campaigns, Product Studio and Group Shot. We tested it head to head before switching, and on the images we deliver it is clearly better, with sharp print, logos and fine text on the default tier. That tier now costs €0.21 an image.

There are four quality levels. Medium at €0.11 is enough for plain garments and simple scenes, high at €0.21 is the default, extra high at €0.32 handles dense pattern, small type and fine knit, and max at €0.63 gives the most detail the model can produce. Teams that already used GPT Image 2 were moved to 2.5 high automatically. Everyone else can switch it on in Campaign Settings, or pick it per perspective in Product Studio.

Find our full research in this blog post.

Your set, your light, your crop

We first built this for Breuninger's marketplace spec in July. If you don't sell on Breuninger, that email probably didn't mean much to you. But almost every brand has a spec of its own, even if nobody ever wrote it down: how much floor sits under the shoes, which side the light comes from, where the shadow falls.

One of our customers, a Danish brand, wanted a very particular look. Their own studio, soft light from one side, and the shadow falling the same way on every image. That is almost impossible to put into words. Even with a model that follows instructions well, the shadow jumped from left to right and the light did something different on every render. With our most careful prompting, somewhere between half and three quarters of the images came out right. You can live with that on a campaign. On a catalog you can't, because nobody can afford to throw away every second or fourth image across a thousand products.

Now each perspective takes a visual guide: a photo of your own empty set. The model is rendered into that set, and the light and shadow follow it. On that setup, roughly nine images in ten now come out right. That simply wasn't possible before.

After the render, the file gets finished the way a retoucher would finish it, by measuring. You can pin the model between two lines, with the top of the head a set number of pixels from the top edge and the soles a set distance from the bottom. A packshot gets the same margin on all four sides. Files come out at an exact pixel size, and the background lands on the exact hex colour you set. If a pose leaves too little room for the crop, you get the full image back with a "Delivered uncropped" label instead of a bad crop.

And if you do sell on Breuninger, the Rest and Sport formats now join Flat.

Lower prices, and previews stay free

Many of you have never seen our prices, so here they are.

On pay as you go, releasing a product image now costs €5. A campaign image is €25 and a video €10. Previews are free and unlimited, so you can work on a look for as long as you need and pay only when you release it. Full-resolution renders are billed at what the AI provider charges us plus a small running cost, so a finished product image comes to about €6 all in.

On an Enterprise plan, a finished image comes to around €4 or less, all in, with no release fee.

Also new

The Quality-Assurance-Agent now takes instructions per perspective. It already scores every Product Studio render against your product photo. You can now tell it what "correct" means for a specific view, such as "the sole must be fully visible", or switch scoring off where you don't need it.

Perspectives can be grouped into tabs, for example one for your marketplace pack and one for campaign views, and dragged into the order you work in. Your whole team sees the same tabs.

The Brandmachine agent answers about twice as fast. The wait used to be noticeable. Now it feels like a conversation.

If AI images haven't made it into production for you yet, reply and tell me what stopped you.