Why This Product Can't Use Subject 2's Synchronous Pipeline

A new engineer, familiar with text-to-image-diffusion-models's synchronous inference chain (prompt safety -> enhancement -> generation -> harm detection -> super-resolution, completing in single-digit-to- low-double-digit seconds), proposes reusing that same synchronous request/response pattern for the headshot product: the user uploads photos, and the API call returns finished headshots directly.

1. Explain specifically why this doesn't work for this product, using concrete numbers.
2. Describe the asynchronous pipeline this subject proposes instead, stage by stage.
3. Which single stage in the async pipeline should stay synchronous, and why?