Mito, a company working to democratize AI-powered video creation, used a serverless, asynchronous architecture on Google Cloud to deliver a 120-second national TV spot for Movistar Plus+ (part of Telefónica), broadcast across Spain during the World Cup. The production team describes the campaign as 95% AI-generated, with 1,441 assets live on a single shared Canvas near launch.

The project exposed the bottlenecks that appear when moving from a single five-second clip to a cohesive campaign. For Movistar alone, production required generating an “iceberg” of over 5,000 assets — 1,000 of them videos — across roughly 65 iteration rounds. Local media libraries on individual workstations did not scale, feedback loops depended on manual downloads, Google Drive uploads, and fragmented chat and email, and standard AI video rendering took four to five minutes per short clip, freezing the UI when run synchronously. Every generative model also has distinct prompting logic, which made character persistence, lighting, and camera angles hard to hold across dozens of shots while preserving the likeness of real public figures such as chef Ferran Adrià and journalist Juanma Castaño. Public pay-as-you-go model APIs added rate limits and regional throughput throttling during peak traffic.

The stack runs the API, the AI Director, and background workers as Java services on Cloud Run, which scales with request volume. Google Cloud Pub/Sub acts as the central event bus to decouple job execution, so long-running video generations are offloaded to background Cloud Run workers and creators keep working on the UI canvas. Media assets are stored in Cloud Storage and delivered via Cloud CDN, with expiring key signatures keeping pre-release client assets private; keeping storage alongside the models called through Vertex AI reduces network transfer delays. An isolated database architecture on Cloud SQL and PostgreSQL handles heavy business intelligence and analytics workloads without risking production performance. Foundational models including Omni, Veo, and Nano Banana are accessed through Vertex AI alongside custom system prompts.

The structured canvas frontend enforces a “Rule of 6” — limiting prompt variables across World, Camera, Light, Texture, Actions, and Character — and prioritizes still-frame generation before video rendering to maximize cost efficiency. The AI Director copilot translates natural-language instructions into precise model parameters while maintaining context, lighting, and visual style across scenes. A multi-model routing layer coordinates tasks across specialized models, including Sonnet, Gemini, and ElevenLabs for audio, treating the spatial canvas like a live code repository to infer adjacent contextual cues.

For enterprise reliability, Mito built a virtual queuing system that mitigates third-party API rate limiting during usage spikes, with bounded retries and a clear outcome for every job. Because creative output cannot be validated with standard unit tests, an automated evaluation harness powered by Gemini acts as an independent judge, scoring AI Director outputs against original user intents across an automated test suite that runs alongside standard tests for high-impact Director changes.

The article also notes boundaries and direction. The production deliberately embraced an “imperfect aesthetic”: in the spot, a 17-year-old student named Tommy makes an AI video project, and at the climax the AI misunderstands a prompt, making coach Luis de la Fuente say “In the same milkshake” while holding a drink; Tommy edits the prompt live to “In the same heartbeat” and the video recalibrates instantly. Surreal visual elements — stadium stands built into Galician cliffs, a neighborhood bar covered in pitch grass — were meant to show generative AI elevating human storytelling rather than obscuring it. Mito runs an internal studio, Mito Films, to dogfood the tools on high-stakes client projects, and the architecture is positioned for rendering times to collapse from minutes to seconds as world models and deeper memory states arrive.