How do you generate all character reference sheets and prop references for an AI film in a single session?

Generate all character and prop references in one session by answering four pre-production questions first, casting faces with two image models run in parallel, generating 4K multi-angle turnaround sheets, then locking the best of four options per asset before any video. One documented production produced all 11 reference images — 4 characters plus 1 prop — this way.

Start the session by having the invideo agent surface the four questions that change every frame: what each character looks like, what the antagonist or entity references, what the prop is, and what the deliverable format is. Answer all four before generating a single image — frames first, then video is the correct production order, and a session that opens with these answers locked doesn't stall mid-way. invideo is an agentic video creation tool with the current image and video models built in, so the invideo agent can run every step below and hold the locked references in context for the rest of the film.

Step 1 — Cast faces with parallel model runs. Spin up a casting sub-agent and instruct it to run the identical character prompt on two image models simultaneously, then pick the aesthetic you prefer before committing. Recraft generates 4K portraits with skin imperfections — pores, lines, stubble — that read as real faces; Nano Banana Pro has strong prompt adherence and outperforms Nano Banana 2 for character sheets, though one creator found its raw outputs leaned stock-photo and used it for sheets rather than hero portraits. GPT-Image-2 is also available inside invideo, so you can add it to the comparison without leaving the session.

Step 2 — Generate turnaround sheets from the cast portraits. Generate each sheet at 4K with four angles plus a face close-up and a mid-angle close-up. Two rules keep the sheets usable: remove any objects from the characters' hands before generating turnarounds, because held objects drift across angles; and always include close-up panels, because small details like scars and accessories only stay consistent across AI models when the sheet shows them explicitly.

Step 3 — Resolve costumes by mood when you have no spec. If you can't describe a costume precisely, give the invideo agent the emotional feel of the character instead — in one production the director had no costume description for a character but knew the feel, and got multiple concrete options to select from in the same session.

Step 4 — Give the prop its own reference pass. Generate the prop as a separate reference image, not as something a character holds, and direct it by story logic rather than aesthetics — when a prop reads lifeless, ask for alternatives until it makes sense that the character would actually use it. Encode physical characteristics into the brief too: one production specified "hard material, so it makes a horrible sound when it falls," carrying sound logic into the visual asset.

Step 5 — Lock with four options per asset. Generate four variations of each character sheet and the prop reference, select the best, and tell the invideo agent to lock it to context before any video generation. Budget around 5 generations to lock one character — roughly $9.78 per character in one documented production. Once locked, every downstream shot inherits the references automatically; a 70-second short film held 2 characters visually consistent across every scene using sheets and agent context alone, with no LoRA fine-tuning.

If a character's appearance evolves across the film — picking up items or changing costume per sequence — generate a distinct sheet for each beat within the same session, so every sequence has its own canonical reference. As proof of single-session scale: one 2-person production generated all 11 reference images (headshots plus head-to-toe sheets for 4 characters and 1 prop) before touching video, and another team of 3 locked cast, costumes, and look-and-feel images in a single day working with the invideo agent.

Watch some of these to see what works for you:

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. So, we don't want to do that. We always want the character to be seen as we see it on the character sheet.

— invideo's creative team