by Cora Rodrigues
As a strategist, I’ve been building decks for over a decade, and it’s always been the most challenging part of my job. Research? You develop an intuition for it—there’s a methodology behind turning data into compelling stories. Structuring those stories into an intuitive slide flow that captivates and convinces? That becomes easier with practice too. Words, after all, are something strategists can usually handle.
But when you’re not a designer, making those slides look good? That’s the real challenge.
The Stock Image problem
I’d spend hours browsing Behance and Dribbble for references, then struggle to apply our branding (thankfully, GLADTOBE has an awesome identity that allows for creativity) while balancing text with visuals. The real pain point was always images—the Achilles’ heel of every deck.
Endless scrolling through Unsplash and Pexels, settling for “good-but-not-great” stock photos that kind of match what you’re trying to portray. Rarely ideal, and there’s always that weird moment when you spot the same image in someone else’s campaign.
Then came Sora and we stopped hunting for images. Started creating them.
Our first Sora experiment
We started cautiously, experimenting with prompts and adding film grain for realism. Our first Sora-powered deck was for an OOH campaign.
(By the way, I’m Cora, the person behind most GLADTOBEONTREND decks you see on LinkedIn. Tips are always welcome in the comments!)
The process was surprisingly collaborative. Sora’s interface feels like an open-source prompt library—you can browse trending creations, see the exact prompts used, and even remix others’ work directly.
After experimenting, we discovered that good prompting practices emerge naturally. For our OOH deck, we used a real placement as reference with this straightforward prompt:
The results were impressive on the first few tries. We also figured that when you have a visual reference, you don’t need overly technical prompts. But creating from scratch requires much more detail—you’re essentially briefing a machine, and your communication skills matter.
Tackling complex visual challenges
Our next challenge was bigger: showing how different countries consume media for a GLADTOBEONTREND piece. We needed to demonstrate GLADTOBE’s full-funnel capabilities across Radio, CTV, and Podcasts in cohesive visuals.
Generic channel imagery looks cheap (anyone who’s searched for license-free TV pictures knows this struggle), and slapping our logo on screens makes it worse.
The solution we found was to showcase our branded merch in local settings and let Sora cast “models” using these tricks:
- Reference makes things much easier. A single concrete image plus a concise prompt often outperforms a wall of descriptive text.
- Negative prompts are your friend. Adding specific “avoid” cues (extra fingers, plastic skin) dramatically improves human realism.
- Consistency = photoshoot vibe. Re-using camera specs, lighting styles, and grain keeps a series cohesive.
Radio and podcasts came easily; television needed extra scene-blocking so the living room didn’t morph into a cinema with mutant couches.
Another tip: when you need to improve your prompt or make it more detailed, use Google AI Studios. It usually refines and polishes it better than those generic prompt listings you see around.
Remember: AI doesn’t understand our reality. It doesn’t know how living rooms work or how people typically arrange themselves on couches. But it seems one LLM can help you communicate with another fellow LLM.
Why we’re sharing this
Generative media isn’t our core product at GLADTOBE; it’s a creative approach we deploy when a visual can amplify the story. This post isn’t a how-to guide—design pros have far deeper prompt engineering knowledge on Midjourney, VEO3, and so on.
AI isn’t just about making decks look better – it’s about reclaiming creative time. Those hours used to spend hunting through stock photos can be invested in refining strategy and storytelling.
What’s next
While we’re having fun with visual content, the real transformation is happening in how we approach client work. We’re building AI workflows that handle the manual stuff – using LLMs for data scraping and reporting, creating agents for SOSTAC-based marketing strategies, and optimizing case study generation.
The learning curve is real, and honestly, we’re still figuring it out as we go. For a nice deeper AI integration, check out Eric’s article on using AI to optimize TV campaign timing.
