What pre-production questions should you ask an AI agent before generating any film assets?

Before you generate a single asset, force the invideo agent to answer four pre-production questions: what does the protagonist look like (character spec), what is the antagonist or entity reference, what is the prop specification, and what is the deliverable format — frames first or straight to video. invideo's creative director calls these the four things that "will change every frame."

invideo is an agentic video creation tool that runs a crew of sub-agents (creative producer, storyboard, DOP, costume, production designer) on top of every current image and video model — so the answers you give to these questions get loaded as project context once and then carried into every downstream generation. Treat the questions as a creative brief: lock identity constants up front, leave action variables (camera move, blocking, lighting shift) for per-shot direction.

1. Who is the character, and what locks down their appearance? Pin down look, era, wardrobe direction, and any small continuity details (scars, accessories, a specific necklace) before any video gen. The standard the invideo agent works to is comprehensive character sheets — front, side, back, and close-ups — because "the AI always needs to see what the character is exactly, right? Or else it'll kind of hallucinate and imagine something that's under the cap." One documented production locked 4 characters and 1 prop with 11 reference images and ~5 generations per character (~$9.78 each) before touching video. Across documented productions, two characters held consistently for a 70-second short with no LoRA, purely off character sheets plus agent context.

2. Who or what is the antagonist / entity, and what is the reference? Give the agent a named visual reference for the threat, love interest, or any second character — "closer to Bathsheba?" is the kind of specificity that unblocks generation. Without this, the agent has to guess the silhouette, era, and tone of every shot the entity appears in. If the entity evolves across the film (a trinket added each sequence, a costume change at the midpoint), say so now — that's the trigger to generate a separate character sheet per beat rather than one master sheet.

3. What is the prop specification, including diegetic logic? Props get their own pass, not a footnote. Specify material, scale, sound logic, and the narrative role — "hard material, so it makes a horrible sound when it falls" is a real prompt input that shapes both the image and how the prop is shot. Generate 4 options per prop, pick one, and lock it; lifeless props break believability even when the character render is perfect. Strip props out of characters' hands before generating multi-angle character sheets so the turnaround stays consistent.

4. What is the deliverable, frames-first or straight to video? Decide the output shape now — static frames approved to quality, then video; or direct-to-video with reference images. Frames-first is the safer default because you lock visual consistency before motion compounds the errors. Also specify aspect ratio in your film's delivery format, runtime target, and which model the agent should route to (the invideo agent holds Runway, Veo, Kling, and Seedance 2.0 — Seedance 2.0 reference-to-video is the usual pick when you need character and location context carried across clips).

A few questions worth adding once these four are answered: what is the visual style reference (upload the frames or stills you want the agent to internalize, and tell it explicitly what to take and what to ignore); what are the non-negotiable continuity rules (lighting source language, palette, lens grammar); and what is the emotional register or stage structure of the film, so the agent can self-check shots against it later. One documented horror short encoded five emotional stages with locked camera, lighting, and sound rules per stage — the agent then flagged a reveal shot running at the wrong stage register that the director had missed.

The payoff for spending real time here shows up downstream: documented productions that answered these questions thoroughly hit $315–$1,500 per project across 2–5 day timelines, with one 2-minute promo coming in at ~$1,500 versus a $100,000–$500,000 traditional equivalent. As Hridaye, invideo's creative director, puts it: "the more clarity you bring to the project, the more sharply the invideo agent will hold it for you across the project."

Watch some of these to see what works for you:

Before I build assets, four things will change every frame: The Girl: What does she look like? What era? The Entity: Closer to Bathsheba? The Toy: Doll, ball, something else? The Deliverable: The frames first, then video? These four answers unlock everything.

— Hridaye, invideo's creative director