If you came looking for Dream Machine, it is worth saying plainly that the product under that name is gone. Luma retired it: lumalabs.ai/dream-machine now returns a 308 permanent redirect to lumalabs.ai/app, and no page on the marketing site uses the term any more. This is not a rebrand of a standalone clip generator into a new standalone clip generator - the shape of the product changed. What exists today is Luma, described by the company as a creative partner, working across image, video, audio and text on a single infinite canvas. Luma's own video model is Ray 3.2. It handles text-to-video and image-to-video at 5 or 10 seconds, with 1080p as the resolution ceiling, and adds a set of editing operations that matter more than the raw generation: Modify Video rewrites existing footage for up to 20 seconds at 24fps, Reframe converts aspect ratio for up to 12 seconds, and the model supports motion transfer between clips, character transformation, environment changes, relighting and product swapping. Shot control runs to 16 keyframes per clip, and Expressive Facial Performance covers up to 8 faces in a single shot.
Try this alternative. The specification that separates it from the rest of the category is colour: Ray 3.2 generates native 16-bit colour and exports EXR in ACES2065-1, which means output can enter a professional grading pipeline rather than arriving pre-baked. One real limitation: text-to-video and image-to-video generate no audio at all. Modify Video and Reframe preserve whatever audio the source had, but a generated shot is silent, and that is a live disadvantage against Veo 3, which produces dialogue and effects natively. The second change is that Luma now runs other companies' models alongside its own. Inside the app you can reach Veo 3.1, Kling 3.0 and Kling Omni, Nano Banana and Nano Banana Pro, Seedream, Seedance 2.0, GPT Image 2, and ElevenLabs for music, sound effects and voice. That is a strong argument for the subscription in its own right, because comparing three video models on the same prompt would otherwise mean holding three separate accounts - and it partly answers the silent-video problem, since ElevenLabs audio is a step away rather than an export away
Explore this category.