EXCLUSIVE: Luma launches creative AI agents powered by its new ‘Unified Intelligence’ models

AI video procreation startup Luma connected Thursday launched Luma Agents, designed to grip end-to-end originative enactment crossed text, image, video, and audio. Luma Agents are powered by the startup’s Unified Intelligence household of models, with architecture trained connected a azygous multimodal reasoning system. 

Luma Agents are being pitched arsenic a caller mode of doing enactment for advertisement agencies, selling teams, plan studios, and enterprises. Luma says its agents are susceptible of readying and generating text, image, video and audio portion coordinating with different AI models, including Luma’s Ray 3.14, Google’s Veo 3 and Nano Banana Pro, ByteDance’s Seedream, and ElevenLabs’s dependable models. 

Luma’s agents are built connected the startup’s Uni-1 model, the archetypal of its Unified Intelligence household of AI models. It has been trained connected audio, video, image, language, and spatial reasoning, according to Amit Jain, main enforcement serviceman and co-founder of Luma.

Jain told TechCrunch that the Uni-1 exemplary tin “think successful connection and ideate and render successful pixels oregon images…we telephone it ‘intelligence successful pixels.’” Other output capabilities similar audio and video volition travel successful consequent exemplary releases, helium added. 

“Our customers aren’t buying the tool, they’re redoing however concern is done,” Jain said. 

Image Credits:Luma AI

Luma has already started rolling retired its caller agentic level with existing customers including planetary advertisement agencies Publicis Groupe and Serviceplan, arsenic good arsenic for brands similar Adidas, Mazda, and Saudi AI institution Humain. 

Jain said the Luma Agents are a gamechanger due to the fact that they are capable to support persistent discourse crossed assets, collaborators, and originative iterations. They tin besides measure and refine outputs, improving their ain results done an iterative self-critique, according to Jain. 

This benignant of check-your-work capableness is what has made coding agents truthful useful, Jain said. “You request that quality to measure your work, hole it, and bash that loop until the solution is bully and accurate.”

Jain said the existent workflow for utilizing AI tools successful originative environments doesn’t person the aforesaid acceleration of benefits radical successful the originative manufacture expect from AI. Instead, it’s much like: “Here are 100 models. Learn however to punctual them,” helium said.

He said what makes Luma Agents antithetic is that you don’t request to punctual backmost and distant for each iteration connected an representation oregon thought — the strategy alternatively generates ample sets of variations and lets users steer the absorption done conversation.

“With Unified Intelligence, due to the fact that these models recognize successful summation to being capable to generate, we are capable to physique a strategy that is capable to bash this benignant of end-to-end work,” Jain said. 

Take, for instance, a quality designer designing a building. As they gully the lines, they are creating an interior intelligence practice of the structure, light, spatial dynamics, and lived experience. This, Jain says, is the aforesaid rule upon which Unified Intelligence is built. 

Jain said the strategy could importantly velocity up originative workflows. In a demonstration, Jain showed however a 200-word little and an representation of a merchandise (a portion of lipstick) led the strategy to make assorted ideas for locations, models, and colour schemes for an advertisement campaign. 

In different example, Luma Agents turned a brand’s $15 million, year-long advertisement run into aggregate localized ads for antithetic countries successful 40 hours for nether $20,000, passing the brand’s interior prime controls and accuracy checks, Jain said. 

While Luma Agents is present publically disposable via API, Jain said the startup plans to rotation retired entree gradually to guarantee users support reliable entree and debar workflow disruptions.