Exclusive: Luma AI Debuts Creative Agents Fueled by 'Unified Intelligence' Models

On Thursday, AI video generation startup Luma introduced Luma Agents, a system built to manage complete creative workflows spanning text, images, video, and audio. These agents are driven by Luma’s Unified Intelligence model family, featuring an architecture trained as a single multimodal reasoning engine.

Luma is positioning its agents as a transformative tool for advertising agencies, marketing departments, design studios, and enterprises. According to Luma, the agents can plan and produce text, images, video, and audio, while also coordinating with other AI models such as Luma’s Ray 3.14, Google’s Veo 3, Nano Banana Pro, ByteDance’s Seedream, and ElevenLabs’ voice models.

The foundation of Luma’s agents is the company’s Uni-1 model, the inaugural member of its Unified Intelligence AI model series. Amit Jain, CEO and co-founder of Luma, explained that Uni-1 has been trained across audio, video, imagery, language, and spatial reasoning.

Jain shared with TechCrunch that the Uni-1 model can "think using language and visualize and render in pixels or images…we refer to this as 'intelligence in pixels.'" He noted that additional output formats, such as audio and video, will be introduced in future model versions.

“Our clients aren’t just purchasing a tool—they’re reimagining their entire business process,” Jain stated.

Image Credits:Luma AI

Luma has already begun deploying its new agentic platform with current clients, which include global ad firms Publicis Groupe and Serviceplan, as well as brands like Adidas, Mazda, and the Saudi AI firm Humain.

Jain emphasized that Luma Agents represent a breakthrough because they preserve ongoing context across assets, team members, and creative revisions. He added that the agents can assess and enhance their outputs, refining results through repeated self-evaluation.

This self-checking ability, Jain noted, is what has made coding agents so effective. “You need that capacity to review your work, correct it, and repeat the cycle until the outcome meets quality and accuracy standards.”

Jain observed that the current approach to using AI tools in creative settings doesn’t deliver the productivity gains the industry anticipates. Instead, it often feels like: “Here are 100 models—now learn how to prompt each one.”

He explained that Luma Agents stand apart because users aren’t required to repeatedly prompt for each iteration of an image or concept. Instead, the system generates extensive variations and allows users to guide the creative direction through conversational input.

“Thanks to Unified Intelligence, our models not only generate but also comprehend, enabling us to construct a system capable of managing complete end-to-end tasks,” Jain said.

He drew a comparison to a human architect designing a building. As the architect sketches, they form a mental representation of the structure, lighting, spatial relationships, and real-world experience. Jain explained that Unified Intelligence operates on a similar principle.

According to Jain, the system can dramatically accelerate creative processes. In one demo, he illustrated how a 200-word brief and a product image (a lipstick) enabled the system to produce multiple concepts for ad campaign settings, models, and color palettes.

In another case, Luma Agents converted a brand’s year-long, $15 million ad campaign into multiple localized advertisements for various countries—accomplishing the task in 40 hours for under $20,000, while meeting the brand’s internal quality and accuracy standards.

Although Luma Agents is now accessible via a public API, Jain mentioned that the startup intends to phase in access gradually to ensure dependable service and prevent workflow interruptions.