How do you produce an AI fight sequence without choreography expertise?

You produce an AI fight sequence by delegating the choreography: screen-record a fight from a film you admire and upload it to the invideo agent to dissect the moves, storyboard the beats, lock a character sheet per fighter per fight state, then generate and stitch the clips. One documented solo production ran this with 5 parallel sub-agents.

The invideo agent is an agentic video creation tool with the current video models (Seedance 2.0, Kling, Veo) built in, and it lets you spin up specialized sub-agents that share one project context — that shared crew is what replaces choreography expertise here.

1. Borrow the choreography from a reference film. Screen-record a fight sequence you admire and upload the recording to the invideo agent with an instruction to study the choreography. The agent dissects the moves, spacing, and rhythm, then reinterprets them for your characters and world — you never have to describe a single strike yourself. One documented episodic production sourced its entire fight choreography this way. If you have no reference, describe the beat in plain language and let a choreography brainstorming sub-agent propose the move-by-move breakdown instead.

2. Storyboard the beats before spending video credits. Have a storyboard sub-agent convert the dissected choreography into panels — one documented production generated 3 storyboard grids for a single fight sequence before rendering any video. Annotate every panel with camera movement, subject action, and overall mood so each frame is machine-readable when it goes to video generation. A storyboard costs image credits only, and video credits are the expensive ones.

3. Lock fighter identity per state. Build separate character sheets for before, during, and after the fight — bruises, torn costumes, and dirt each count as a new state, and a single static sheet will not carry across them. Give every fighter, including the antagonist, their own named sheet: distinct visual references stop the model from blending fighters or cloning your hero into both roles — an error one documented production fixed by generating separate sheets per opponent, after which the second generation attempt succeeded without further intervention.

4. Run the fight as parallel sub-agents. Split the work into discrete tasks and give each one its own sub-agent. One documented production deployed 5 sub-agents on a single fight sequence: choreography brainstorming, context updating, image generation, image-to-video conversion, and an alternate version of the same fight developed in parallel. Because every sub-agent inherits the same project context — character sheets, location references, format — you never re-explain anything between tasks, and the alternate version gives you a second cut to choose from at no extra planning cost.

5. Generate with action-friendly camera direction. Route the clips to Seedance 2.0 for contact-heavy action, and specify camera handling in the prompt — for example 'war zone documentary style' — so shaky, chaotic movement reads as realistic combat without demanding frame-perfect shot coherence. Kling is cited by some filmmakers for slow-motion beats; every roster model runs inside invideo, so the invideo agent routes each shot rather than you picking a platform per model. Budget roughly 3 generations per usable shot.

6. Stitch the sequence. The invideo agent can generate a slate — it stitches your approved fight clips into one complete final sequence automatically. Where no single generation nails a full shot, build a Frankenstein shot: combine the best seconds from 2 or more generations of the same prompt; in one documented episode, 17 of the final shots were composites.

Watch some of these to see what works for you:

I screen recorded that section, and I put it on Agent One, and I asked it to study the choreography of it. And it surprisingly dissected it ironically without even using a knife.

— a solo AI filmmaker documenting an agent-directed fight sequence