Photos and video made with AI, without a photoshoot

A website about schools needs pictures of schools: classrooms, teachers, students, devices in hand. Stock photos give themselves away at once, because they are the same everywhere. A photoshoot across Italian schools would have meant weeks of work, permissions and release forms.

For Kyron we took another route. We generated the images with AI, under clear rules, and checked every result.

The homepage video: ten seconds on a loop

kyronedu.it opens on a girl sitting on the floor against a wall. She reads on an iPad and, halfway through, takes a sip of coffee. It's a ten-second clip generated with Seedance from a still image.

The hard part was colour. AI never hits an exact shade, and the wall had to be Kyron's deep teal. We solved it in three steps: we recoloured the starting image, generated the motion, then corrected only the tones of the wall. Skin, floor and backpack stayed as they were.

The clip ends almost exactly where it begins, so it loops without a visible cut. The whole file is under two megabytes.

Page photos, with two fixed rules

The photos on the other pages were generated with Higgsfield and Nano Banana. Kyron gave us two rules, and every image follows them:

• European people only;
• no text inside the image, not on screens and not on signs.

On top of that comes the style: soft light, real-looking rooms, a touch of Kyron's green. We wrote it all down as a fixed recipe, so every new photo looks like the ones before it, even months later.

AI still gets some details wrong, so we check them one by one. A tablet propped at an angle looked like it was floating: we redid it lying flat on the desk. A laptop came out with an impossible shape: redone as well.

When a product has no photo

Out of about 80 products in the catalogue, six had no photo: furniture, labs on wheels, things that are hard to shoot. Instead of a grey box we generated an abstract mark in Kyron's green, which stands in until the real photo arrives.

The same method produced the kit images in the school shops and the picture for the Kyron Shield protection plan: an iPad, a shield, no words.

Fix, don't redo

Sometimes a photo is almost right. Then we don't generate it again from scratch: we edit it, and change only the part that doesn't work.

On the support page, for example, the big screen showed an abstract gradient. Kyron wanted something believable. We changed only the screen, filling it with charts and dashboard cards, and left the people, the room and the light untouched.

Two launch videos written in code

Two Kyron products, a compass for funding calls and the search in Kyron Space, needed launch videos. Generative AI wasn't the right tool here: they had to show the real interface, the exact colours and the brand's fonts.

We built them entirely in code, with Remotion. Every frame is drawn by the program: no camera, no generation. The payoff comes later. To change a word or a colour you edit one line and render again, in minutes.

What changed

Kyron has a consistent look on every page without having organised a single photoshoot. When a new image is needed, it follows the same rules and the same style as the rest.

For video, the choice depends on the job. Mood and people are generated with AI. Interfaces and precise messages are drawn in code.