On Thursday, the AI video-generation company Luma introduced Luma Agents, a system built to manage complete creative projects involving text, images, video, and audio. These agents are driven by the company's Unified Intelligence model family, which uses an architecture trained on a singular multimodal reasoning system.

Luma is presenting its agents as a transformative tool for advertising agencies, marketing departments, design studios, and large corporations. The platform can plan and produce content across text, image, video, and audio formats, and it coordinates with other AI models such as Luma’s own Ray 3.14, Google’s Veo 3 and Nano Banana Pro, ByteDance’s Seedream, and voice models from ElevenLabs.

The foundation of Luma Agents is the Uni-1 model, the inaugural member of the Unified Intelligence family. According to Amit Jain, Luma's CEO and co-founder, this model has been trained on audio, video, images, language, and spatial reasoning. He noted that additional output capabilities for audio and video will be introduced in future model updates.

Jain emphasized that the product represents a fundamental shift in operations, stating, “Our customers aren’t buying the tool; they’re redoing how business is done.”

The new agentic platform has already begun deployment with several existing clients. These include major global ad agencies like Publicis Groupe and Serviceplan, as well as brands such as Adidas, Mazda, and the Saudi AI firm Humain.

Jain described Luma Agents as a breakthrough due to their ability to maintain consistent context across different assets, team members, and creative versions. The agents can also assess and polish their outputs, enhancing results through an iterative self-review process. Jain likened this critical self-evaluation capability to what has made coding agents so effective, adding, “You need that ability to evaluate your work, fix it, and do that loop until the solution is good and accurate.”

He contrasted this with the current experience of using AI in creative fields, which often lacks the expected efficiency gains. The typical process, he said, is more akin to being presented with numerous models and told to learn how to prompt each one. Luma Agents differ by eliminating the need for repetitive prompting for each iteration; instead, the system produces large batches of variations and allows users to guide the direction through conversational feedback.

“With Unified Intelligence, because these models understand in addition to being able to generate, we are able to build a system that is able to do this sort of end-to-end work,” Jain explained. He drew a parallel to an architect designing a building, who develops an internal mental model encompassing structure, light, space, and experience—a principle he says is central to Unified Intelligence.

The system promises to greatly accelerate creative workflows. In one demonstration, Jain showed how a 200-word brief and a product image (a lipstick tube) led the system to generate numerous ideas for ad campaign settings, models, and color palettes.

In a more substantial example, he reported that Luma Agents converted a brand’s $15 million, year-long advertising campaign into multiple localized ads for different countries. This process was completed in 40 hours for under $20,000 and successfully met the brand's internal standards for quality and accuracy.

While Luma Agents is now accessible to the public via an API, Jain stated that the company plans to expand access in a controlled manner. This phased approach is intended to ensure reliable service for users and prevent disruptions to their workflows.