Luma Launches AI Agents for Creative Workflows — Built on a New Unified Intelligence Architecture That Thinks, Imagines, and Renders as One
Deployed today with global enterprise partners including Publicis Groupe and Serviceplan Group, Luma Agents introduce a new class of AI collaborators that execute end-to-end creative work across text, image, video, and audio — without fragmenting context or requiring manual orchestration between disconnected tools.
5 min read
Luma has announced the launch of Luma Agents, a new class of AI collaborators capable of executing end-to-end creative work across text, image, video, and audio. Designed for agencies, marketing teams, studios, and enterprise organisations that need to scale creative output without sacrificing quality, Luma Agents maintain full context from initial brief to final delivery — coordinating tools, models, and iterations within a single unified system.
The launch marks a significant architectural departure from how most AI creative systems have been built. Rather than chaining together separate models for language, vision, video, and reasoning and stitching their outputs via orchestration layers, Luma has built its Agents on a new foundation it calls Unified Intelligence — a single multimodal reasoning system trained to understand and generate across formats within the same architecture.
"Creative work has never lacked ambition; it's lacked execution capacity. 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."
— Amit Jain, Co-Founder and CEO, Luma
The Problem with Fragmented AI Creative Pipelines
For the past several years, the dominant approach to AI-powered creative work has been to assemble pipelines: one model writes text, another generates images, another processes video, and orchestration layers attempt to stitch their outputs together. While effective for narrow, isolated tasks, these systems suffer from a structural limitation — they fragment context between steps, require complex workflow management, and produce inconsistent results when multiple creative directions need to run in parallel.
The practical consequence for creative teams is significant: switching between disconnected tools, manually rebuilding context at each handoff, and absorbing the cognitive overhead of orchestration rather than focusing on creative direction. Luma's position is that intelligence should not be assembled in pieces — it should be built as one coherent system.
What Luma Agents Can Do
Luma Agents replace fragmented, multi-model workflows with coordinated execution built on unified reasoning. They operate inside a collaborative, multiplayer environment where humans direct creative intent and Agents handle orchestration, routing, and execution. Specifically, Agents are designed to:
- Execute projects end-to-end — from planning through production and delivery — without requiring manual handoffs between tools
- Maintain shared context across text, image, video, and audio throughout the entire creative lifecycle
- Advance multiple creative directions in parallel, enabling teams to explore more options without multiplying workload
- Evaluate and refine outputs iteratively through self-critique, rather than producing single-pass, one-shot results
- Integrate into enterprise tools and production systems via API, fitting into existing workflows rather than requiring wholesale replacement
Agents also coordinate across a wide set of leading AI models — including Ray3.14, Veo 3, Sora 2, Kling 2.6, Nano Banana Pro, Seedream, GPT Image 1.5, and ElevenLabs — automatically selecting and routing tasks to the best model or capability for each specific step, and maintaining persistent context across assets, collaborators, and creative iterations.
The Architecture Behind It: Unified Intelligence and Uni-1
Luma Agents are built on Unified Intelligence, a new model architecture that takes a fundamentally different approach to multimodal AI. Rather than connecting specialised models after the fact, Unified Intelligence trains a single multimodal reasoning system capable of understanding and generating across formats — text, image, video, and audio — within the same architecture. The design principle is that reasoning and rendering should be tightly coupled, not separated into sequential steps.
Luma draws an analogy to human creative cognition: when an architect sketches a building, they are simultaneously simulating structure, light, spatial dynamics, and lived experience — reasoning and imagination happen together, not in sequence. Unified Intelligence is built on the same principle.
The first model built on this architecture is Uni-1 — a decoder-only autoregressive transformer operating over a shared token space that interleaves language and image tokens, allowing both modalities to function as first-class inputs and outputs within the same sequence. This design enables the model to reason in language while imagining and rendering in pixels within the same forward pass, producing creative outputs as part of a single coherent reasoning process rather than across disconnected systems.
"Intelligence shouldn't be fragmented by modality. Unified systems reason holistically. When the same model can think, imagine, and render, you move closer to intelligence that behaves coherently across the entire creative process."
— Amit Jain, Co-Founder and CEO, Luma
Already Deployed at Global Scale
Luma Agents are not a preview or beta — they are already embedded across the operations of major global agency partners. Publicis Groupe and Serviceplan Group are deploying Luma Agents across strategy, creative development, and production workflows to increase throughput while maintaining brand consistency across markets.
"Luma is now part of our broader House of AI ecosystem and integrated directly into our creative workflows. 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."
— Alexander Schill, Global CCO, Serviceplan Group
Enterprise-Ready Safeguards
Luma Agents are designed with enterprise IP protection, compliance, and operational scale as core requirements — not afterthoughts. Key safeguards built into the platform include full IP ownership retained by customers, automated content review to reduce copyright risk, legal trace documentation demonstrating human involvement, required human review workflows prior to public release, and cloud-based infrastructure with enterprise-grade guardrails throughout.
The combination of Unified Intelligence architecture, broad model coordination, and enterprise-grade controls positions Luma Agents as a genuinely distinct category — not isolated generation tools, but collaborative AI creatives capable of executing the full scope of creative work from brief to delivery.
Key Takeaways
- Luma has launched Luma Agents — a new class of AI collaborators that execute end-to-end creative work across text, image, video, and audio, already deployed with global enterprise partners Publicis Groupe and Serviceplan Group across more than 20 countries
- Agents are built on Unified Intelligence, a new architecture that trains a single multimodal reasoning system rather than chaining together separate models — tightly coupling reasoning and rendering so the system can plan, imagine, and produce within one coherent process
- The first model on this architecture, Uni-1, is a decoder-only autoregressive transformer that interleaves language and image tokens in a shared token space — enabling the model to reason in language while rendering in pixels within the same forward pass
- Luma Agents coordinate across leading AI models including Ray3.14, Veo 3, Sora 2, Kling 2.6, GPT Image 1.5, and ElevenLabs — automatically routing tasks to the best model at each step while maintaining persistent context across assets and collaborators
- Enterprise safeguards include full customer IP ownership, automated copyright risk review, legal trace documentation of human involvement, required human review prior to public release, and enterprise-grade cloud infrastructure