Design Sora (Text-to-Video)
Hard · System Design · about 45 minutes
Asked at OpenAI, Google, Meta, Adobe, NVIDIA, ByteDance, Unity.
Design a text-to-video service like Sora. Users submit a prompt and options (length, resolution) and receive a generated video minutes later. Each job occupies one or more GPUs for 1–10 minutes, demand regularly exceeds the fixed GPU fleet, free/paid/enterprise users expect different service, and GPU failures are routine. Prompts and outputs must pass safety checks.