Why it matters
The story we keep hearing is that generative AI levels the creative playing field - anyone with a prompt can now make a video.
The KRO-NCRV experiments show something more nuanced. The same 60 seconds of audio produced three radically different videos depending on who was at the keyboard. Generative tools don't eliminate the gap between novice and expert; they relocate it. And in this case the audience mattered: KRO-NCRV asked us to imagine deaf and hearing-impaired children, which sharpened every choice. We tracked the costs too - 8 to 18 hours, €66 to €79, 66 to 138 kg of CO₂. Knowing all of this matters for anyone making decisions about learning time, hiring budgets, or trust.
How we did it
We kept the input identical and varied the maker.
The same 60-second podcast segment about the first internet connection in the Netherlands was given to three set-ups: a novice in Runway Gen-3 Alpha with default settings (#01 Starter Kit); a prompt engineer using Runway's advanced suite (#02 Made an Effort); and a filmmaker with Midjourney, Adobe Premiere, and full AI expertise (#03 Make It Rain). Each was time-boxed, costed, and carbon-tracked. Every choice was filtered through one question: would this work for a deaf or hearing-impaired child watching the video?
What we made
Experiment #01 - Starter Kit
Transcript-led text-to-video generation in Runway Gen-3 Alpha, executed by a complete novice. 8 hours of work, €74 in tool credits, 72 kg of CO₂. The baseline experiment: what happens when anyone with default settings tries to turn audio into video.
Experiment #02 - Made an Effort
Text-to-video with prompt engineering expertise, using Runway Gen-3 Alpha + Alpha Turbo. 12 hours, €66, 66 kg of CO₂. A creative with generative AI fluency working with the full Runway suite and advanced settings - to see what one level of expertise actually changes.
Experiment #03 - Make It Rain
Full filmmaker workflow: text-to-image (Midjourney) + image-to-video (Runway) + Adobe Premiere editing. 18 hours, €79, 138 kg of CO₂. The high-end experiment: a filmmaker with animation, editing, and AI expertise pulling out the stops.
A complete write-up of what we learned across the three experiments.
Side-by-side comparison of approach, time, cost, carbon, and quality. The dossier covers what to expect when scaling generative AI workflows from novice to expert, where the bottlenecks live, and what specific trade-offs the tools force you to make.
The full open archive of the project
Research notes, episode source materials, experiment specs, BTS screenshots, and ongoing community contributions.
Research Insights
Side-by-side comparison of approach, time, cost, carbon, and quality. The dossier covers what to expect when scaling generative AI workflows from novice to expert, where the bottlenecks live, and what specific trade-offs the tools force you to make.
More projects.
(2019-25©)
