Canvas-Native AI Agents Bring Spatial Context to Generative Workflows
Luma AI published details on Laya, its architecture for canvas-native AI agents that analyze spatial layouts, notes, and asset groups inside visual workspaces. Creative directors and storyboard artists gain an assistant that understands surrounding context instead of isolated text prompts.
Luma AI detailed its framework for canvas-native creative agents with Laya, shifting generative workflows from traditional text-box prompting to interactive spatial boards. The system reads visual layouts, text notes, asset groupings, and image references simultaneously to assist creators directly within an infinite workspace canvas.
What's new
Traditional AI video platforms rely on isolated prompt inputs, forcing creators to re-specify context, framing, and visual style for every generation. Luma Dream Machine's underlying research into Laya addresses this by giving AI agents continuous spatial awareness across an entire workspace board.
The canvas-native architecture processes the relational context of items on a board. If an editor places mood board stills next to generated video variations, Laya recognizes proximity, reads accompanying text notes, and tracks version history across visual frames. Rather than executing single text requests, the agent observes shot sequences, structural groups, and visual notes to assist with iteration, asset sorting, and multi-shot continuity directly on the visual grid.
How it fits your workflow
For filmmakers, animators, and visual effects directors, canvas-native interaction replaces rigid prompt boxes with a visual environment similar to Figma or Miro. Creators can layout storyboards, mood boards, and character design references on a shared canvas, letting the AI agent generate variations that respect surrounding visual notes.
This spatial approach addresses continuity challenges common in standard generative workflows. When building scene transitions or multi-angle sequences, creators can arrange keyframes side-by-side. The Laya agent interprets positional context to keep character models, color palettes, and lighting consistent across generated clips. This workflow positions Luma AI against prompt-centric platforms like Runway Gen-3 Alpha, ComfyUI, and Pika 1.5, offering a visual workspace model for creative asset development.
What it costs / how to try it
Luma AI outlined the canvas-native architecture and lessons from Laya on its engineering blog as of February 2025. Access to canvas-native agent features is rolling out iteratively within Luma's experimental feature sets and Luma Dream Machine platform accounts.
Read the original announcement on Luma Dream Machine ↗