Automated Concept Ideation + Storyboarding → Google Slides
Mar 2025
In 2024, there were a few significant LLM advancements that came about in a short time period:
Contextual reasoning
Image generation capabilities
Deep research was introduced.
It was for this reason that I put together this proof of concept in roughly one week. Using scripts and API calls, it ideates and generates contextually aligned concepts and storyboards for video creative. The concepts and visuals are then automatically assembled into Google Slides for easy updating and optimization.
Most AI tools generate generic, de-contextualised ideas. This workflow bakes in context from the start, ensuring outputs are on-brand, consistent, and immediately useful for creative development.
Note: This initial POC is context-driven (not fully branded). I later expanded the approach into branded ideation using an agentic workflow (orchestration that incorporates brand books, tone, and constraints).
Put simply, it's a context aware creative-concept generator
What it does
Ingests deep research – Utilizes Deep Research outputs and/or textual context, converting into creative concepts.
Creates concepts w/ reasoning – Generates campaign concepts and provides reasoning behind creative choices.
Builds the storyboard – Drafts scene text, voiceover, and detailed image prompts for each shot.
Generates visuals – Uses Vertex AI Imagen to create images, which are stored in Google Cloud Storage.
Builds the deck – Takes a templated Slides deck, fills it with copy and images, and creates various narratives.
Supports character consistency – Implements a degree of character continuity through prompt engineering across scenes to reduce visual drift.
How it works
Extracting Slides template structure – The process begins by converting a prepared Google Slides template into a JSON outline via a custom GCS script. This gives a clear map of slide layouts and placeholders to work with.
Adding context – Details are provided to the LLM regarding what the user wants to create -so the AI isn’t just generating generic content but ideas based on a given request and context.
Storyboard generation – Using that context, the AI produces the concept itself: scene text, suggested dialogue/voiceover, and detailed image prompts.
Visual creation & storage – Image prompts are turned into visuals using LLM API calls, then uploaded to Google Cloud Storage. Each slide gets a direct link to its generated artwork.
Slides automation – Finally, the system populates the template and fills it with the generated text and images, keeping everything in sequence. Notes, background images, and extra details can also be embedded for each slide.
