Leadership 2026, an AI short film
AI video and motion design for a three-minute animated short, how one of Power's senior leaders came to the company and rose through it, 90 Wan 2.2 generations in my own ComfyUI pipeline, played live as he spoke on stage.
- Year
- 2026
- Campaign
- Leadership
- Client
- Power Home Remodeling
- Role
- Senior motion designer
- Discipline
- generative, motion
- Tools
- ComfyUI, Wan 2.2, Qwen Image Edit, RIFE, After Effects
- Turnaround
- TurnaroundOnly where it is impressive
One of Power’s senior leaders wanted a film for Leadership 2026 about how he came to the company and rose through it, told as a cartoon, with another senior leader drawn in beside him. He supplied the frames, generated in ChatGPT, with the character designed with help from his family. I turned them into motion.
Every clip runs from one of his frames to the next, generated on my own machine with Wan 2.2 image to video in ComfyUI. That’s what lets three minutes read as one continuous story. It runs from a street corner through a bar, a gas station, a job site and a whiteboard that says process is king, to a wave from a pickup beside a Window Pros truck.
The graph is ComfyUI’s Wan 2.2 first-last-frame setup, tuned for this job. Both ends of every clip are pinned to a supplied frame, so the model only invents the motion between them, and pinning the same frame at both ends gives a loop with no visible join that he could hold on while he talked. Wan 2.2 splits the work between two 14 billion parameter experts, one that lays out the motion in the early noisy steps and one that resolves the detail, and Lightning LoRAs cut the pass to four steps so a take came back fast enough to judge and run again. RIFE doubles the frames at the end. It was my first large-scale AI video, and today I’d run it on other models with a local Llama rewriting the prompts, but it’s where the pipeline started.
That speed is what the job needed. It took 90 video generations and 11 image edits to get three minutes I’d put on a stage. The first transition alone took eleven takes, and the prompt was rewritten after the third. Where his frames didn’t fit the cut, I made new ones with Qwen Image Edit, rain on the security guard, the truck flipped to drive the other way with its logo intact, an empty window for it to drive into.
It started from his script and went through several rounds of revision, and it shipped as a set of clips he cued himself while he spoke on stage. The two of them liked it enough to send me a bottle of bourbon.
Final
Brief
Boards
Styleframes
Build
Delivery
Stills
- StillExports
- StillExports
- StillExports
Breakdown
The graph behind every clip. Both ends pinned to a supplied frame, two experts splitting four steps, RIFE doubling the frames
Storyboards
The frames he supplied, generated in ChatGPT, in story order. Every clip runs from one of these to the next
Styleframes
- StyleframesProcess material
- StyleframesProcess material
- StyleframesProcess material
In progress
Made the guard look left and brought in rain, puddles and grey pants, so the cut back to him lands in a new scene
The Window Pros truck flipped to drive the other way, with the logo kept the right way round
The pickup and the character taken out, leaving an empty view for the truck to drive into
The whiteboard wiped and the arms lowered, a clean start for process is king to be written
Loops
Eight of the eleven takes of the first transition, the prompt rewritten after the third
Eight of the transitions, each one frame into the next
The street corner into the bar
From the job site to a customer's door
The framing into the crew
Process is king
The pickup and the wave