Building a narrative film with Generative AI
An experiment using the latest generative image and video models to translate high bandwidth memory concepts into a coherent visual story. The goal was to test how far an AI only pipeline could go while we retained creative direction and technical judgment.
Overview
I built a 34 Frame storyboard, matched models to each production task, tested continuity shot by shot, and curated viable outputs. The result was approximately 90% generative, with creative industry tools required for final motion and graphic finishing.
Production Mix
90% AI Generative Workflow / 10% Creative Post-Production
Models Selected By Task
ChatGPT
Claude + Connectors + MCP
Internal Product Specifications
FLUX1.1 [pro] Ultra Raw
FLUX.1 Kontext [max]
FLUX.2 [pro]
Gemini 3.1
Firefly Image Model 5
Luma Ray 3.14
Veo 3.1
Firefly Video
Kling 3.0
Runway Gen-4.5
Topaz Bloom
Adobe After Effects
Adobe Photoshop
Creative Process
How the Sequence Was Built
One sequence mapped from reference and frame development through motion generation and finish.
Build the Visual Language
Reference + Geometry → Visual Development → Motion
Data Flow Visual
Use translucent blue and cyan cubes to represent data movement and bandwidth.
Product Geometry
Strengthen the eight HBM packages with a reflective titanium finish and clear, bold labels. Preserve the board layout and surrounding components.
Angle Refinement
Build the Data Interaction
Guide the cubes toward the HBM packages, then shift to a low 45-degree isometric view while keeping the hardware geometry and materials consistent.
Data Absorption
Animate the cubes streaming into the HBM packages and resolve to a clean board. Keep the camera stable and the chip layout intact.
Design the Camera View Transition
Start + End frames → Controlled Camera Move → Finish
Isometric View Board
Lock the clean 45-degree isometric composition as the opening frame.
HBM Close Up
End on a macro view of a single titanium HBM package.
Controlled Zoom In
Continue from the isometric frame with a controlled move toward one HBM package, matching the supplied end frame.
Post Production
Assemble the sequence, correct continuity, refine timing and typography, and upscale the final output.
Topaz Bloom
After Effects
Photoshop
What I learned so far. AI accelerated design production, but achieving full 1:1 accuracy and maintaining visual continuity remained key challenges.
The harder design problem was keeping product geometry, material behavior, and technical details consistent across scene transitions. These limitations may narrow as AI creative platforms integrate more capable models, stronger reference controls, custom training, and grounded workflows!
For enterprise creative teams, quality alone is not enough. Creative direction, brand governance, product accuracy, rights management, and control over proprietary IP determine whether generated work is usable.
The opportunity is not simply faster generation, but a governed, AI-assisted workflow in which human judgment guides the system, evaluates outputs, and delivers a controlled and brand aligned result.
Date: Feb 2026