Generative AI in Ad Studio

The generative features I designed and shipped in Ad Studio during 2025, one at a time. It started with Product Photography on Flux.AI: isolate the product from a creative, then place it in a new scene described in the chat prompt. Gemini 2.5 Flash and reference images came next, then video with Sora 2 and Veo 3.1, including first-and-last-frame clips. The platform’s earlier text and image tools, from 2024, were another team’s work.

Type
AI / UX
Role
Senior Product Designer, sole designer
Team
1 PM, 1 engineer, 1 QA, 1 ML engineer
Year
2025
Product Photography in the Ad Studio editor: a brand creative on the canvas with the Product photography panel open on the right
Every new feature landed in the editor, on the same panel as the rest of the image controls.
Where it started: Flux.AI
The Product Photography flow across its states: start, image uploaded, and using an image as a mask, each shown with the editor and the panel
Isolate the product from the creative, describe a new scene in the chat prompt, and Flux.AI renders it there. Click to zoom in and pan across it.

The product comes straight from the creative, so it stays exactly the same

An upgrade: Gemini 2.5 Flash, aka Nano Banana
The same Product Photography flow across two model providers, Flux Kontext and then Gemini 2.5 Flash as the default
Same interface, better engine: the default model swapped from Flux Kontext to Gemini 2.5 Flash, the one the public came to know as Nano Banana. Click to zoom in.

The panel was built for the model underneath it to change

An addition: reference images
The Select assets modal: choosing reference images from Brand Kit folders, with the option to upload new ones, up to ten
Steer the result with up to 10 reference images, pulled from the Brand Kit or uploaded fresh. Click to zoom in.
A new medium: video
The video panel with the model picker open: Sora 2 and Veo 3.1 options, each with an explainer describing what the model is good for, from text-to-video to first-and-last-frame and reference-guided clips
Then video, with a range of models behind one panel. Sora 2 for text-to-video and image-to-video; Veo 3.1 for first-and-last-frame motion and reference-guided clips, each with an explainer for what it’s good for. Click to zoom in.
The Ad Studio editor with a brand creative selected on the canvas and the Image to video panel open, model set to Sora 2
Image-to-video, straight from a creative already on the stage. Click to zoom in.

Each new model and feature landed in the same editor, so nobody had to relearn the workflow

Looking for a senior who can own it?

If that's the kind of person you're hiring, let's talk.

Copies hello@tamaskiraly.com to your clipboard
hello@tamaskiraly.comlinkedin.com/in/kiralytamasProfessional career