Prompt as Screenplay: Directing with Language
A text prompt and a generated video work like a screenplay and a finished film. Both need vision, specificity, and the willingness to be surprised by what comes out.
When I write a prompt for Sora, I'm not typing instructions. I'm directing. I'm making choices about light, mood, the weight of a moment. The language shapes what appears, not mechanically, but interpretively. The AI reads my words the way an actor reads a script: with its own understanding, its own biases, its own surprising choices.
I've started thinking of prompts as screenplays. Not structurally (there's no three-act arc in ten seconds) but intentionally. A good prompt, like a good screenplay, is specific where it matters and open where it doesn't. It gives the system enough to work with and enough room to surprise you.
The best films I've generated came from prompts that described a feeling, not a scene. "A woman alone in a city that doesn't notice her" produces something more alive than "a woman walking down a busy street at night." The first is direction. The second is description.
This is what I mean by conscious creation with AI. You're not giving up authorship. You're expanding it. You're learning a new language of collaboration where the medium talks back.