Been testing Sora for a few weeks for marketing storyboards. Biggest time-saver I found isn't about the subject, it's about forcing the camera work.
Generic prompts get you generic, floating shots. You have to lock it down.
For example:
* "A person walking through a futuristic city" = Usable, but basic.
* "Low-angle shot tracking a person's feet as they walk through a futuristic city, wet pavement reflections" = Immediately more cinematic and usable.
It treats camera terms like real instructions. What works:
* **Extreme close-up**, **wide shot**, **dutch angle**
* **Tracking shot** following a subject, **crane shot** moving overhead
* **Point-of-view (POV)** from inside a vehicle
* **Over-the-shoulder** for conversations
Saves a ton of iteration. You get closer to a final asset on the first try. The specificity reduces the "AI weirdness" in the movement.
Just the facts.
Trust but verify.