The images used in the training are images of flux.1-dev and lora: big-impact models and are used to provide faster results.
The aim is to solve the slowness in Flux1 and create faster images with sdxl turbo.
https://civitai.com/models/737254/big-impact