Luma has unveiled a new class of “creative AI agents” built on its freshly announced Unified Intelligence model architecture, promising end‑to‑end campaign execution across text, images, video and audio for agencies, brands and studios worldwide. The Palo Alto based startup says the system is already being deployed at global creative networks to boost output and speed without sacrificing craft, positioning Luma as an emerging infrastructure layer for multimodal creative work.
Luma Agents are described by the company as AI “collaborators” designed to take projects from brief to finished assets, rather than standalone generation tools that require manual orchestration at every step. They can plan and generate copy, images, video and audio, while coordinating with external specialist models from companies such as Google, ByteDance and ElevenLabs.
According to Luma, the agents maintain persistent context across assets, iterations and collaborators, allowing teams to adjust direction conversationally instead of writing long prompt chains for each revision. In demonstrations, the company showed agents turning a 200‑word creative brief and a single reference image into a range of ad campaign ideas and asset variations, and localising a reported 15‑million‑dollar campaign for multiple markets in roughly 40 hours.
Amit Jain, co‑founder and CEO of Luma, framed the launch as an attempt to resolve a structural bottleneck in creative production rather than to replace human teams. “Creative work has never lacked ambition; it’s lacked execution capacity,” Jain said in a statement. “Creative teams shouldn’t have to spend their time orchestrating tools. They should spend it creating. Agents aren’t shortcuts. They’re collaborators that maintain context, coordinate execution, and advance projects so teams can focus on taste, direction, and strategy.”
Underpinning the new agents is Luma’s Unified Intelligence architecture, which the company positions as an alternative to the prevailing industry practice of chaining together separate language, vision and generation models. Rather than assembling “intelligence in pieces,” Luma trains a single multimodal reasoning system intended to understand and generate across formats within one coherent model.
The first model in this family, Uni‑1, is a decoder‑only autoregressive transformer that operates over a shared token space interleaving language and image tokens. This design allows the model to reason in natural language while “imagining and rendering in pixels” inside the same forward pass, effectively treating both modalities as first‑class inputs and outputs. Jain told TechCrunch that Uni‑1 can “think in language and imagine and render in pixels or images … we call it ‘intelligence in pixels,’” with richer audio and video capabilities slated for subsequent versions.
Luma argues that this coupling of reasoning and generation brings the system closer to how human creatives work, where planning and visualisation happen together rather than in discrete stages. “When a human architect sketches a building, they are not simply drawing lines , they are simultaneously simulating structure, light, spatial dynamics, and lived experience,” the company’s technical overview notes, adding that Unified Intelligence is built on the same principle.
On top of this base, Luma Agents can coordinate complex workflows that previously required several tools and extensive manual coordination. The company says the agents can automatically select and route tasks to the most suitable model, maintain global context across campaigns and teams, and run iterative self‑critique loops to evaluate and refine outputs until they meet predefined criteria.
The launch comes amid a broader shift in generative AI from single‑shot models toward agentic systems that can manage multi‑step tasks. Luma positions its agents as a replacement for fragmented multi‑model workflows that have emerged as creative teams experimented with a growing ecosystem of AI tools.
Today, many agencies and in‑house teams toggle between different text, image and video models, rebuilding context and re‑entering briefs at each stage. Luma says its agents instead keep a persistent understanding of the campaign, brand guidelines and creative direction, allowing them to “remember” previous decisions and carry them forward into subsequent iterations and formats.
Jain told TechCrunch that one key differentiator is how users interact with the system as ideas evolve. Rather than prompting “back and forth for each iteration on an image or idea,” Luma’s approach is to have the agent generate large sets of variations while users steer direction conversationally. “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,” he said.
A separate company write‑up emphasises that Uni‑1 “grows a mind’s eye from a logical brain,” describing it as unified reasoning and visual imagination in a single model capable of maintaining temporal and spatial consistency while evolving scenes. That capability is intended to support complex video and 3D outputs as the platform matures.
Luma says its agents are already being used across large agency networks and creative groups globally, although the company has not disclosed the full list of customers. Publicis Groupe and Serviceplan Group are among the early adopters deploying Luma Agents across strategy, creative development and production workflows to increase throughput while preserving brand consistency across markets.
“Luma is now part of our broader House of AI ecosystem and integrated directly into our creative workflows,” said Alexander Schill, global chief creative officer at Serviceplan Group. “It allows our teams across more than 20 countries to collaborate more smoothly and develop great work faster. For our clients, that means high‑quality creative output delivered with greater speed and efficiency without compromising craft.”
According to Luma, agents operate inside a multiplayer, collaborative environment where humans set creative intent and constraints while the AI handles orchestration, routing and execution. The company claims this setup can increase creative “velocity” and output volume while giving teams a single space to coordinate campaigns that span multiple formats and markets.
Beyond agencies, Luma is targeting enterprise marketing departments, entertainment studios and global brands that need to scale asset production across languages, channels and geographies. The company says its Unified Intelligence platform is already serving teams at leading advertising agencies, global enterprises and entertainment studios, though details remain limited.
While Luma’s own models sit at the core of the system, the company is leaning into an orchestration role that embraces third‑party tools rather than attempting to replace them. The agents can coordinate across external AI models including Luma’s Ray 3.14, Google’s Veo 3 and Nano Banana Pro, ByteDance’s Seedream, GPT‑based image models and ElevenLabs for voice.
In practice, that means the agent can choose the best specialist model for a particular task for instance, a high‑fidelity video generator for cinematic shots, or a particular voice model for a regional campaign and route prompts and context accordingly. Users interact at the level of campaign objectives and brand guidelines, while the agent manages which underlying systems to call and how to align their outputs.
The agents’ self‑evaluation loop, which Luma says is central to its offering, aims to reduce the need for manual quality control on every generated asset. By automatically testing outputs against defined criteria and regenerating when necessary, the system is designed to move closer to the “review and refine” cycle that human teams perform, albeit under human supervision.
The launch of Luma Agents follows a period of rapid growth for the startup, as demand for AI solutions in media, advertising and entertainment continues to accelerate. Luma, headquartered in Palo Alto, has spent the past several years building unified multimodal AI systems that combine reasoning and generation, initially gaining attention for its video and 3D capabilities.
Last November, the company raised around 900 million dollars in funding, with plans to develop large‑scale “super” data infrastructure in Saudi Arabia, according to recent coverage of its expansion. Investors include Humain, a subsidiary of Saudi Arabia’s Public Investment Fund, along with AWS, AMD Ventures, Nvidia, Amplify Partners and Matrix Partners, among others.
That capital has allowed Luma to double down on its Unified Intelligence roadmap and pursue partnerships with major creative and media players. By framing its agents as infrastructure for creative work rather than standalone consumer tools, Luma is implicitly competing with both foundational model providers and production‑focused startups that offer more specialised services.
Industry observers see agentic systems as a next phase of enterprise AI adoption, shifting from experimentation with single models to integrated platforms that can take on complex workflows. Luma’s bet is that a single, unified reasoning and generation core will ultimately outperform manually stitched‑together pipelines in both quality and efficiency.
For creative professionals, the arrival of systems like Luma Agents raises a now‑familiar mix of opportunity and concern. On one hand, agents that can maintain context, version assets and coordinate across channels promise to offload repetitive production work and free humans to focus on concept, narrative and brand voice. On the other, deep integration into agency workflows could reshape team structures and expectations around speed, output volume and billable work.
Luma’s leadership has taken pains to describe the agents as collaborators rather than replacements. “Agents aren’t shortcuts,” Jain reiterated in the company’s launch materials. “They’re collaborators that maintain context, coordinate execution, and advance projects so teams can focus on taste, direction, and strategy.”
Early adopters echo that framing, at least publicly. Serviceplan’s Schill emphasised that the goal is to deliver “high‑quality creative output” faster and at scale “without compromising craft,” describing Luma as one component in a broader ecosystem rather than a wholesale replacement for existing tools or roles.
As with other generative AI deployments, issues such as data governance, IP, bias and transparency will remain central as the technology rolls out more broadly. Luma has stressed the importance of unified reasoning and consistent context as a way to keep outputs coherent and aligned with brand standards, but has shared fewer details publicly on dataset composition, rights management or safeguards around synthetic media misuse.
For now, the launch of Luma Agents underscores how quickly the generative AI landscape is moving beyond single‑model tools into systems that promise to run entire creative processes. Whether Unified Intelligence and Uni‑1 become a new standard for multimodal agents or one of several competing approaches, the company’s latest move signals that the race to define the infrastructure of AI‑driven creativity is entering a new phase.
Comments