Luma Dream Machine

Luma Dream Machine was one of the first AI video generators people actually talked about. If you have come looking for it, the honest answer is that Luma Labs has built something larger around it.

The Dream Machine address now leads to Luma, headlined "creative agents for creative professionals" — an infinite-canvas workspace covering images, video, audio and text rather than a box you type a video prompt into.

This review covers what replaced the standalone generator, why running other companies' models is the interesting decision, and who the new shape actually suits.

Why This Listing Changed

Last checked 7 September 2026. Checked directly against the Dream Machine address, with the screenshot above captured at the same time.

• lumalabs.ai/dream-machine now serves a page headlined "Creative agents for creative professionals" under the name Luma
• The model list — Ray 3.2, Ray 3.14, Uni-1, Veo 3.1, Kling 3.0, Kling Omni, Seedream, Seedance 2.0, GPT Image 2, Nano Banana and Nano Banana Pro — read from Luma's own page
• Capabilities including animation, style transformation, voiceover, lip sync and localisation as Luma lists them
• Free trial confirmed on the page; commercial use is attached to a paid plan per Luma's own description
• Pricing, API and enterprise sections confirmed present in the site navigation
• No plan prices published here, in line with this site's pricing policy

This listing is kept under the Dream Machine name because that is what people still search for. The product itself is now Luma.

What Happened To Dream Machine

• The name has moved on. lumalabs.ai/dream-machine now presents Luma — "creative agents for creative professionals" — not a standalone video generator.
• What it is now: an infinite-canvas workspace where agents plan, generate, iterate and refine across images, video, audio and text.
• It runs many models, not only Luma's own — Ray alongside Veo, Kling, Seedream, Seedance, GPT Image and Nano Banana.
• Beyond generation: animation, style transformation, voiceover, lip sync and localisation in the same canvas.
• Free to try, with commercial use attached to a paid tier.

Luma Workspace Details

From One Generator To A Workspace

Dream Machine did one thing: text or image in, video out. That was remarkable in its moment and has since become table stakes — a dozen products generate competent video from a prompt.

Luma's answer was to stop competing on the generation step and compete on everything around it. The pitch now is keeping full context across a project: plan, generate, iterate, refine, without losing the thread between stages.

That is a different product for a different buyer. A generator serves someone who wants a clip; a canvas serves someone producing a body of work where shot three has to match shot one.

Generating one good clip stopped being the hard part. Producing forty that belong together never stopped being hard.

Running Everyone Else's Models Too

The most striking thing about the current Luma is the model list. Alongside its own Ray 3.2, Ray 3.14 and Uni-1, it offers Veo 3.1, Kling 3.0 and Kling Omni, Seedream, Seedance 2.0, GPT Image 2, and Nano Banana and Nano Banana Pro.

A model developer offering competitors' models inside its own product is a deliberate bet: that the workspace, not the model, is where the durable value sits.

For users it removes a real problem. Model quality leapfrogs every few months, and being locked to one vendor's release cycle means your output quality is hostage to their roadmap. Picking per shot is strictly better.

What The Canvas Adds

The infinite canvas is not decoration — it is how the context is kept.

• Everything in one place — images, video, audio and text side by side rather than in four tools
• Iteration stays visible — earlier versions sit next to the current one instead of vanishing into history
• Animation and style transformation applied to work already on the canvas
• Voiceover and lip sync, so audio is not a separate export-and-reimport step
• Localisation for versions in other languages
• Agents that carry out multi-step work rather than answering one prompt at a time

Judging It Against Its Former Self

If you specifically wanted Dream Machine's simplicity, the new shape will feel like more product than you asked for. A canvas and an agent layer are overhead when you want one clip.

If you were frustrated by that simplicity — regenerating endlessly, losing versions, stitching together output from four tools — the change is squarely in your favour.

The practical test is volume. One clip a month, and a simple generator wins. A project's worth of related material, and the workspace earns its keep quickly.

Getting Oriented Quickly

Coming from a prompt-box generator, the canvas takes some adjusting to. This order avoids the usual frustration.

1. Start with a single shot to learn the generation quality before touching the agent features.
2. Try the same prompt across two or three of the available models — the differences are larger than the marketing suggests.
3. Keep rejected versions on the canvas rather than deleting them; comparison is the point of the format.
4. Build one short sequence before attempting anything long, so continuity problems surface early.
5. Add audio and lip sync only once the visuals are settled, since revisions invalidate them.
6. Check the commercial use terms before putting anything into client or published work.

Honest Reservations

Breadth carries a cost. A workspace spanning images, video, audio, text, animation, lip sync and localisation is a lot of surface, and products this wide are rarely best-in-class at any single point.

Model choice cuts both ways too. Offering eleven models is powerful and also a decision you now have to make repeatedly, with little guidance on which suits what.

Commercial use sitting on a paid tier is worth noticing before you build anything client-facing on the free trial.

And the rename leaves a trail. A large amount of writing about Dream Machine describes a product that no longer exists in that form, which makes research on it unusually confusing.