Mastering AI Video Marketing w/ Magnific CEO Joaquín Cuenca Abela | AI Basics

What It Covers

Joaquin Cuenca Abela, CEO of Magnific (formerly Freepik), demonstrates how a single non-creative person can produce a cinematic AI video ad in 24 hours using text prompts, storyboards, and layered media references — covering production costs, model architecture, Hollywood adoption, and the future of hyper-targeted dynamic video advertising.

Key Questions Answered

• •AI Video Production Cost: Magnific charges approximately $0.10 per second of generated video, but accounting for multiple iterations to achieve the desired output, founders should budget roughly $1 per finished second. A launch video that previously cost $50,000–$150,000 from a production company now costs a fraction of that figure.
• •Storyboard-to-Video Workflow: Start with text prompts to generate character and scene images, arrange them as storyboards, then convert sequences into video clips. Reference images pulled from the web can supplement prompts when text descriptions lack precision. One non-designer can complete a full cinematic short in a single 24-hour session.
• •Model Stack Strategy: Rather than building every model in-house, Magnific combines proprietary tools — its upscaler and skin enhancer — with best-in-class third-party models from Google and others. Founders building AI creative tools should prioritize integrating state-of-the-art external models rather than reinventing foundational infrastructure from scratch.
• •Hollywood Adoption Pattern: Major studios, including the production behind House of David, now use Freepik/Magnific in production workflows. AI reduces film budgets enough to greenlight previously cost-prohibitive projects — one director cited employing 600 people on a film that would have been rejected entirely without AI-driven cost reduction.
• •Hyper-Targeted Video Localization: Static ad localization across 50 languages and regional visual swaps (food, landmarks) already works reliably at scale. Full video localization — 100 city-specific versions of one ad — remains unreliable due to compounding error rates across generation steps, requiring validation layers before production deployment.