AI Video Essentials

How do you prompt AI video tools like a film director?

Last updated June 26, 2026

Prompt AI video tools the way you'd talk to your DOP on set: give directorial intent in on-set vocabulary — shot type, blocking, lens, motivated lighting, emotional register — not technical parameter strings. Load a treatment doc once so style holds across every shot, then direct shot by shot in plain crew language and let an agent crew route each call to the right model.

Start by writing the way a director briefs a crew, not the way a user prompts a model. That means on-set nouns and verbs: "hold on him, no cutting, push in until he lunges," "reverse on Marcus — the near wall doesn't exist yet, what should it be?", "warm yellow from the lamps only, like all the refs." These are the lines that came back as usable shots in documented productions, and they work because they tell the agent what the scene is FOR, not what dials to turn.

The invideo agent is an agentic video creation platform that holds project context across every shot and routes each generation to the right model (Runway, Veo, Kling, Seedance 2.0) — so directorial language is the input layer, model choice is handled for you.

Use a 5-element prompt structure per shot. Every shot prompt should carry: (1) subject and blocking — who is in frame, where they are, what they do; (2) shot type and lens — master, coverage, OTS, POV, push-in, 35mm spherical; (3) motivated lighting — the source and why ("warm yellow from the lamps only"); (4) atmosphere and palette — mood register, color script, grain; (5) emotional tone or stage — what the audience should feel at this beat. This is the assembly order documented productions used across every frame to prevent drift.

Load a directorial reference once, then stop re-specifying style. Build a treatment document — visual language, camera grammar, lighting rules, palette modes with hex values, what to frame, what to withhold — and upload it to a creative producer agent at project start with the full script. One documented 70-second short loaded a 25-page treatment and ran every shot against 12 parameters (film reference, shot design, length, lens, lighting plan, color script, atmosphere, blocking, emotional register, final prompt, negative prompt, revision prompt) without re-prompting style. As Hridaye, invideo's creative director, put it: "camera continuity carries from the treatment doc forward. you're not telling the agent how to move the camera every time. you set it once. it holds. that's the flow state."

Direct in the crew metaphor — assign typed agents per role. Spin up a creative producer agent first to hold script, shot breakdown, and characters as the central vision. Then assign a storyboard agent to visualize shots before you direct them, a DOP agent (or several — one per scene, since each scene needs a different eye) for camera and lighting, a costume designer agent that responds to mood when you don't have exact spec ("I knew the feel I wanted for her costume" produced multiple viable options), a production designer agent, and a director's assistant agent to sequence the shot breakdown. Documented productions ran 6–8 such agents in parallel.

Talk to each agent the way you'd talk to that crew member on set. To the DOP agent: "I want to stay on the feral guy when we run this scene. No back and forth cutting. We hold on him right up till he lunges." To the costume agent: the feel, not the spec. To the storyboard agent: the beat, not the frame description. When intent is ambiguous, a well-briefed agent surfaces options instead of guessing — "reverse on Marcus, what's behind him? That near wall doesn't exist yet. What should it be?" Treat that as art direction, pick, move on.

Direct the model choice through the agent, not around it. Different shots want different models — Kling for native multi-shot sequences, Seedance 2.0 reference-to-video for character/location continuity across clips, Veo and Runway where their strengths fit — and invideo holds all of them, so the invideo agent routes each shot rather than you swapping platforms. State the intent ("continuous handheld follow, character carrying prop, dead-earth landscape") and let the routing happen underneath.

The end-to-end flow. Outline → shot list (the agent generates this from the script and treatment) → one prompt per shot in your crew-language structure → parallel agent execution → review the rough cut by sending it back to the invideo agent with "what's working, what's not" — documented productions caught wrong emotional-stage registers and pacing errors this way that human editors missed. Work act by act rather than across the whole film so context stays sharp.

The shift underneath all of this: "The thing that made it possible wasn't prompting. It was directing." If you've directed on set, your vocabulary already works — write the brief, let the crew of agents execute.

Watch some of these to see what works for you:

Watch the invideo agent take director-style briefs to crack impossible shots
Day 1: how to pre-produce a short film by directing AI, not prompting it
Six agents, one director: briefing AI crew the way you'd brief humans on set

camera continuity carries from the treatment doc forward. you're not telling the agent how to move the camera every time. you set it once. it holds. that's the flow state.

— Hridaye, invideo's creative director

Share

More on AI Video Essentials