Deploying a text-to-video generation service

Design a video-generation service around a measurable output contract, asynchronous jobs, compatible model inputs, and artifact delivery. Use explicit storage assumptions and measured worker throughput to size the service.

The preceding lesson used Mochi 1 preview to examine video training. Here, use that reference to distinguish the actual model path from optional planning and postprocessing services. The traffic scenario is hypothetical, and the design does not describe a production Mochi service.

Start by separating model artifacts, stored accounts, and daily generated data.

Storage estimation

Assume 100 million daily active users, ten generation requests per active user per day, and separately 100 million stored profiles. Use decimal byte units and distinguish encoded video files from decoded frames.

• Transformer weights: The rounded ten-billion-parameter model at two bytes per parameter occupies: