Generative Vision Interview Questions, issue 17, Jun 25, 2026
The Perceived Noise Paradox
Senior ML Engineer interview at Midjourney, and the interviewer asks:
“You trained a diffusion model that’s gorgeous at 256px. You scale it to 1024px, keep the exact same noise schedule, and the outputs get quietly worse. Same architecture, same loss, same schedule. What broke?”
Don’t say: “The schedule is resolution-agnostic, so I’d just train longer or add more data.”
How zero-mean noise silently destroys your high-res diffusion training, and the timestep shifting trick that rescues your detail-defining steps from being averaged away.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.