Diego Rodriguez

Diego Rodriguez is co-founder and chief technology officer of Krea, the creative AI company he founded with Víctor Pérez. He builds tools that give artists greater control over generated images and video, while challenging whether conventional benchmarks capture visual quality or artistic intent.

Rodriguez studied audiovisual systems engineering at Universitat Pompeu Fabra. He and Pérez, another UPF alumnus, developed the project that became Krea during a period connected with Cornell University and a fellowship from the “la Caixa” Foundation. Their experiments with generative art began before the recent AI boom and included generative adversarial networks. Krea subsequently evolved from standalone creative applications into a platform for generating and editing visual media, training custom styles, and orchestrating multiple models. An Andreessen Horowitz investment announcement put its audience above 20 million users.

• Evaluation grounded in human perception. Rodriguez argues that object counts, prompt adherence, and other convenient metrics cannot reliably distinguish compelling imagery from visual nonsense. His AI Engineer World’s Fair presentation examines how Fréchet Inception Distance can change substantially with JPEG artifacts even when images look almost identical to people. He favors assessments trained on human preferences that recognize artistic intention, including deliberate departures from realism.

• Perceptual compression influences generative AI. JPEG exploits people’s greater sensitivity to brightness than certain color differences; related principles underpin compressed audio and video. Rodriguez applies this background to a neglected evaluation problem: generative models learn from internet media already shaped by perceptual assumptions and compression artifacts.

• Creative control requires aesthetic diversity. Krea’s Krea 2 technical report describes a company-wide effort to balance human aesthetic preferences, prompt fidelity, structural correctness, and stylistic variety while limiting reward hacking. Rodriguez contributed documentation, licensing, and inference-setting changes to the model’s public GitHub repository; the report credits other authors.

Rodriguez also focuses on what happens when generated media becomes abundant: how creative teams find useful material, maintain a distinctive visual direction, and communicate across languages. He has personally used machine translation to support Japanese-speaking Krea customers. His public account of why he founded Krea frames the company around empowering creative people.