How do you run two AI image models simultaneously to compare character casting results?

Run two image models on the same character prompt in parallel — not sequentially — so prompt state, reference attachments, and creative intent stay identical across both. Inside invideo, spin up a casting agent and instruct it to fire the same brief at Recraft and Nano Banana simultaneously, then pick the aesthetic that wins for that character.

Tell the invideo agent to initialize a casting sub-agent, hand it the character description plus any reference images, and instruct it explicitly: run this same prompt on two image models at once and return both results side by side. Parallel beats sequential for three concrete reasons — prompt state stays identical (no silent edits between runs), reference attachments don't drift, and you compare aesthetics in one pass instead of re-running the loop twice.

Pair the models to the job, not the prompt. Recraft is the photoreal portrait specialist — it renders pores, lines, stubble, the skin-level imperfections that make a face read as a real person — so cast it for headshots and any close-up identity work. Nano Banana (use Pro where available) handles full character sheets at 4K with four-angle turnarounds plus face and mid closeups, and accepts multiple reference images for fusion — cast it for head-to-toe references and multi-angle sheets. Run both on the same brief, compare, lock the winner.

Brief the casting agent in one message, not two. Give it the character description, any mood references, and the explicit instruction to test both models in the same round. Where it helps, name the wider model field you want covered — the invideo agent routes to every roster model (Recraft, Nano Banana, GPT-Image-2 for image; Veo, Kling, Seedance 2.0 for video downstream) so you never leave the platform to A/B a model. One production cast 4 characters and 1 prop in 11 total image generations this way, using headshots and head-to-toe references — small image budgets are possible because each round produces real comparisons, not blind re-rolls.

Use mood, not spec, when you don't have one. If you don't have a precise costume or look brief, give the casting agent the feel — "a female vampire, restrained, more menace than glamour" — and it returns multiple concrete options across both models in the same round. Hridaye describes this directly: "I did not have a clear description of the sort of costume… But I always knew the sort of feel I want from her costume. So [the invideo agent] was able to give me multiple options in the same zoom."

Lock the winner with four options, then stop. Once one model's aesthetic wins for a character, ask that model for four variations of the locked sheet and select one. That locked sheet enters the invideo agent's context — it becomes the persistence layer that holds character identity across every downstream shot, no LoRA fine-tuning required. One 70-second short film maintained 2 characters across every scene this way at $750 / 3,000 credits over 2 days.

Generate close-up panels, not just wides. Character sheets must include close-up panels so small details — scars, accessories, jewelry — survive into the video model. And before generating the sheet, strip objects from the character's hands; props in hand break consistency across turnaround angles. If a continuity error appears later, ask the invideo agent to inspect the sheet itself — it identifies the exact panel containing the error, fixes it in place, and stores the corrected sheet so all subsequent shots inherit the fix automatically. Surgical edit, not a re-roll of the whole shot.

These are the moves that make parallel casting a production discipline rather than a tab-switching exercise — what wins for your film depends on the character.

Watch some of these to see what works for you:

I did not have a clear description of the sort of costume for Sylvia, who is our female vampire. But I always knew the sort of feel I want from her costume. So agent 1 was able to give me multiple options in the same zoom.

— Hridaye, invideo's creative director