Luma AI’s agent-driven creative canvas, formerly Dream Machine, built on its Ray3 video models plus third-party image, video, and audio models.
- Own Ray3.2 and Ray3.14 video models alongside Veo 3.1 and Kling 4.0 on one credit balance
- Infinite canvas with Brainstorm mode for planning before spending credits
- Master Reference Assets keep characters and products consistent
- EXR export and team workspaces for production use
- Paid plan required; the pricing page shows no free option
- Entry plan is pricier than most single-model video tools
- Unused credits expire at the end of each billing cycle
Luma AI’s former Dream Machine app has grown into a broader workspace that the company now presents as Luma Agents, and the old Dream Machine address leads to that product. The lab still trains its own Ray video models, but the experience today is an infinite canvas where agents research, generate, and refine images, video, audio, and text side by side.
Luma’s current in-house models include Ray3.2 and Ray3.14 for video and Uni-1 for images. The pricing page also lists third-party engines such as Veo 3.1, Kling 4.0, Seedance 2.5, MiniMax H3, GPT Image 2, Nano Banana, and ElevenLabs audio models, all paid from one credit balance. Workflow features include a Brainstorm mode for planning without spending generations, a Create mode for execution, Master Reference Assets that keep characters and products consistent, lip sync, voiceovers, music, captions, video extension and reframing, plus EXR export for production pipelines.
The workspace is a good match for art directors, brand teams, and storyboard artists who iterate many variations of a concept and want planning, generation, and review in a single shared board. Two practical caveats apply: the pricing page does not list a free plan, with the entry Plus tier at $30 per month, and unused credits expire at the end of each billing cycle. Users who only want an occasional quick clip may find the agent-driven canvas heavier than a simple prompt box.
