So, a few months ago I shared a workflow that generated footage for a single-character ad.
You give it a script. You give it a visual style. It spits out all the shots you need:
Talking head
B-roll
Character-in-action
But here’s the thing... one person talking to camera is fine.
Two people talking to each other? That’s harder to ignore.
It feels more like you’re overhearing something real.
Like, “Oh wait, they’re talking about my problem.”
So I wanted to build that.
Two characters
One script
Full footage
To make this happen… I halted the natural tendency to dive into ‘refining the pompt’ and instead…. did some pre-thinking, even hashed it out on a Miro (for my visual braincells). From there, I split the workflow into phases and flows.
Phase 1: Set the DNA. Captures style, identify settings, world cohesion.
Phase 2: Generate the footage. Separate flows that don’t talk to each other. One flow for Character A. One for Character B. One for both. One for B-roll.
Why separate? Because that’s how you control the chaos. That’s how you debug. That’s how you upgrade one piece without breaking the others.
Let me show you what I mean... (check out the vid breakdown. There’s more to be done, but my 2nd iteration is already heading in the right direction)
