Research & development — III entries

The Lab

The R&D and production systems behind the work. A lot of it comes from directing AI: I write the rules, lock the style, and keep what holds up.

Every entry lists the tools used. Finished client films live in Work.

I.

Production & Pipeline

Systems I've built for American Fidelity's in-house video team: boards, concepts, generative footage and the archive underneath it all.

The Storyboard Machine

A node-based pipeline that produces on-style storyboard frames with the same characters in every shot.

Built in-house at American Fidelity, where we needed to board several competing concepts for a broadcast spot quickly. The flow combines a style reference (a sample of the pencil-sketch treatment) with pools of reference characters — two leads plus supporting cast — so any shot prompt comes out in the same drawing style with the same faces.

The output only reads as real boards because of the shot-writing rules:

  • One frame is one camera view and one moment. Dynamic beats get split across frames.
  • Prompts describe hard scene facts and explicit facial expressions only — no lighting, color or mood language.
  • Every shot prompt stands on its own. Never “same room as before.”
  1. Script
  2. Shot list, written to the grammar
  3. Style referenceCharacter poolsShot prompt
  4. Frame
  5. Board strip
  • Style lock plus character pools make the boards repeatable
  • A written shot grammar turns generative output into usable boards
  • Several concept directions boarded in the time one used to take
Node graphGraphic coming soon
Board strip from a dummy conceptGraphic coming soon

The Concept Room

Developing and stress-testing broadcast spot concepts with AI as a sparring partner.

For American Fidelity’s first-ever televised :30, a single script had been going in circles, so I used Claude as a writers’ room. First I wrote down the real constraints: thirty seconds, a low shot count, no on-screen text the viewer has to read, nothing tied to any one industry, and the actual customer at the centre. Then we generated concept directions and pressure-tested each against that sheet, one pass at a time.

The ideas that survived came with their camera logic built in — locked-off frames, or a single continuous lateral move — so each could go straight to boards.

The same approach covers scripting: demo-driven teleprompter scripts for leadership videos, timed to read naturally in 90–120 seconds.

  • Written constraints first, ideas second
  • Each concept comes with its camera grammar, not just a logline
  • Feeds straight into the storyboard pipeline
Blank constraint sheetGraphic coming soon

Archive Atlas

Turning seven years of raw production storage into a searchable inventory the whole team can use.

American Fidelity’s video archive was a NAS full of year folders, temp holds and project-file dumps, and people only knew where things were by memory. I mapped it into a structured CSV inventory, then brought that into the team’s documentation board so anyone can find a project, its year and its source files without digging.

  1. NAS year folders
  2. Scan
  3. CSV inventory
  4. Documentation board
  • A year-by-year inventory of the full production archive
  • A structured CSV that feeds the team's documentation tool
  • A repeatable pass for each new year's work
Redacted inventory sampleGraphic coming soon

Now playing

Film