What should you load into an AI agent before starting a short film production?

Load six things before you generate a frame: the full script, a visual style document with named references, locked character sheets, world/location plates, your hard constraints (length, shot count, budget, forbidden elements), and a shot breakdown. The invideo agent holds all six in persistent context so every downstream sub-agent — storyboard, DOP, costume, production design — pulls from the same brief.

invideo is an agentic video creation tool where you brief one creative producer agent at the top, and it grounds every specialized sub-agent you spin up after it. So the six inputs below all go into that producer agent first, before any shot, character, or location work begins.

1. The full script (or a tight prose treatment). Upload the complete screenplay in one go so the agent has character arcs, themes, and motifs across the whole film, not scene-by-scene. On a 7-minute animated short the script gets split into three acts so the agent never loses context across a long file. Hridaye, invideo's creative director, frames the setup this way: "To really set up the context for the agent, I normally start off with the creative producer agent. That's where I'll give the script, or the shot breakdown, along with the characters. That's the main agent that sort of holds the understanding and the vision of the entire film."

2. A visual style document (your style bible). This is the highest-leverage upload. Name the reference directors and films, specify lens behavior (spherical vs anamorphic, bokeh shape, flares), lighting grammar (e.g. an 85:15 dark-to-light ratio), color modes with hex values, composition rules, atmosphere, mood register, and a negative-prompt list of what the film must never look like. One documented production used a 25-page treatment with 14 sections covering camera, angles, color tone, atmosphere, mood, lighting, composition, movement, palettes, prompt templates, negative prompts, and a quick-reference card. Another locked an animated style by uploading 64 reference frames in a single message with the instruction to "deeply understand this art style and save it into context for further generations" — plus an explicit prohibition on live-action and photorealistic outputs so the model couldn't drift.

3. Locked character sheets. Generate four options per character, pick one, then build a multi-angle sheet (front, side, profile, back) plus face and mid-angle close-ups at the resolution you'll deliver in. Include close-up panels for small details — scars, accessories, props in hand — because anything not on the sheet, the model will hallucinate. For sequences where a character's look evolves (a costume change, a trinket added each scene), build a separate sheet per beat. One 70-second two-character short hit full cross-scene consistency this way with no LoRA fine-tuning; an Arcane-style episode locked four characters and a prop with 11 image generations and roughly five attempts per character to finalize.

4. World and location plates. Lock the world before you lock shots. Batch your references by theme — spatial logic, screen function, color theory — and tell the agent explicitly what to take from each batch and what to ignore. Generate grids (three per round is a useful default), iterate on the grid you like, then extract the winning panels and use those as the continuity anchors instead of the original references. Once one world element is locked, the invideo agent will autonomously pull wide, close, and side angles off it without you asking for each.

5. Hard constraints. Tell the agent your fixed parameters up front: target film length, rough shot count, deliverable format and aspect ratio for your film, character count, locations, forbidden elements, and your credit/budget envelope. Documented productions ran $750–$5,000 all-in on 2–5 day timelines — the budget shapes how many generations per shot you can afford (planning average ~3 generations per usable shot is realistic). Also load the four pre-production questions the invideo agent asks before generating anything: who the protagonist is, who/what the antagonist is, what the key prop is, and whether you're going frames-first or straight to video.

6. The shot breakdown. A scene-by-scene shot list with the 9-element prompt assembly order baked in — camera spec, lens and aspect ratio, lighting source, palette, composition, atmosphere, mood register, film/DP attribution, negative prompt — so every prompt downstream starts from the same skeleton. Hand the breakdown to a director's assistant sub-agent to sequence shot order before any DOP sub-agent starts generating.

Once those six are in the producer agent, spin up your specialized sub-agents off it — a storyboard agent first to visualize each shot, then a DOP agent (or one per scene for tonal variety), costume, production design, and casting agents. Each inherits the same brief, so feedback to one doesn't cross-contaminate the others. In one documented production six agents ran in parallel; another scaled to eight simultaneous specialist agents across separate project pages and produced a 2-minute brand promo in 3 days that would have taken roughly 2 months on a traditional shoot.

Watch some of these to see what works for you:

To really set up the context for the agent, I normally start off with the creative producer agent. That's where I'll give the script, or the shot breakdown, along with the characters. That's the main agent that sort of holds the understanding and the vision of the entire film.

— Hridaye, invideo's creative director