A hop-kiln office hears that pictures are grown by starting from noise and removing it in stages. Which generative architecture is that?

Denoising diffusion generates images by starting from noise and iteratively removing noise toward structure—exactly the “pictures grown by scrubbing noise in stages” story told in a hop-kiln civic office. That iterative denoising path is the architecture family, distinct from one-shot spreadsheet formulas that claim to print a finished photo in a single cell write. A pure ASR pipeline that only transcribes hop-kiln audio has no image generation path. Architectures that forbid noise and emit only pretrained CSV tables are not diffusion generators. Exam caveat: NCA-GENM wants architecture awareness of diffusion’s iterative denoising, not a step-count bake-off or vendor SKU quiz. Operational check: when reviewing a civic stills generator diagram, identify the noise-to-structure trajectory and the denoiser backbone; do not accept a one-write spreadsheet or ASR-only path as diffusion. Separate diffusion sampling-step discussions from the core claim that generation proceeds by iterative denoising. Diffusion’s identity is iterative denoising from noise to structure; hop-kiln diagrams should show that trajectory, not ASR or spreadsheet one-shots. Sampling-step debates come after that architecture recognition.