We Turned 400 Guests into Forbes Covers in 2 Hours: Behind Our Stable Diffusion AI Stack.
The technical challenges of transforming 400+ guests into a Forbes cover concept at a large-scale corporate event — and how our LoRA, ControlNet, and SDXL infrastructure delivers results in seconds.
- Published
- March 27, 2026
Frequently asked questions.
Have a question on the topic? Write to us and we will reply the same day.
01Which AI model does the AI photo booth run on?
Our stack is built on Stable Diffusion, used together with concept-specific LoRA training and a ControlNet layer that preserves the guest's facial structure.
02How long does a single photo take at a busy event?
At the awards night described in this article the average processing time was 9.4 seconds. What shortens it: model optimisation, warming the system up before doors open and caching the fixed prompts.
03How is facial likeness preserved in an AI photo?
With ControlNet, which takes the face and pose as reference, and IP-Adapter FaceID, which carries the facial identity. Likeness rises markedly with these layers compared with the base model and LoRA alone.
04Where were the photos processed at this event?
At the event described in this article all processing ran on local servers; images were not sent to the cloud and were deleted within the agreed period after the event.