AI video generation startup Luma on Thursday launched Luma Agents, designed to handle end-to-end creative work across text, images, video, and audio. The Luma agent is powered by the startup’s Unified Intelligence family of models, with an architecture trained on a single multimodal inference system.
Luma Agent is being pitched as a new way of working for advertising agencies, marketing teams, design studios, and businesses. Luma says its agents can plan and generate text, images, video, and audio while working with other AI models, including Luma’s Ray 3.14, Google’s Veo 3 and Nano Banana Pro, ByteDance’s Seedream, and Eleven Labs’ audio models.
Luma’s agents are built on the Uni-1 model, the first in the company’s Unified Intelligence family of AI models. Amit Jain, CEO and co-founder of Luma, said Luma is trained in audio, video, images, language, and spatial reasoning.
Jain told TechCrunch that the Uni-1 model can “think in language, imagine in pixels and images, and render…we call it ‘pixel intelligence.'” Other output features such as audio and video will be included in subsequent model releases, it added.
“Our customers aren’t buying tools; they’re reinventing the way they do business,” Jain said.
Luma has already begun rolling out its new agent platform to existing customers such as global advertising agencies Publicis Groupe and Serviceplan, as well as brands such as Adidas, Mazda, and Saudi AI company Humain.
Jain said Luma agents are transformative because they can maintain persistent context across assets, collaborators, and creative iterations. According to Jainism, they can also evaluate and refine their output and improve their own results through iterative self-criticism.
tech crunch event
San Francisco, California
|
October 13-15, 2026
Jain says this type of ability to check work has become very useful for coding agents. “You need that ability to evaluate your work, revise it, and iterate through the loop until the solution is appropriate and accurate.”
Jain said current workflows for using AI tools in creative environments are not accelerating the benefits that people in the creative industries expect from AI. Rather, it’s like, “Here are 100 models. Let’s learn how to prompt the models.”
He said the difference with Luma Agents is that you don’t have to prompt back and forth every time you iterate an image or idea. Instead, the system generates a large amount of variation and allows the user to control direction through conversation.
“With unified intelligence, these models can not only be generated but also understood, allowing us to build systems that can do this kind of end-to-end work,” Jain said.
For example, consider a human architect who designs a building. When they draw lines, they are creating structure, light, spatial dynamics, and internal mental representations of lived experience. This is the same principle on which integrated intelligence is built, Jain says.
Jain said the system can significantly speed up creative workflows. In a demonstration, Jain showed how, based on a 200-word synopsis and a product image (lipstick), the system generates different ideas for locations, models, and color schemes for an advertising campaign.
As another example, Jain said Luma Agents transformed a brand’s $15 million annual advertising campaign into multiple localized ads for different countries in 40 hours and for less than $20,000, passing the brand’s internal quality control and accuracy checks.
Luma Agents are currently publicly available via API, but Jain said the company plans to roll out access in stages to ensure users maintain reliable access and avoid disruptions to their workflows.
