Luma AI launches Dream Machine
Free-to-use video model generating five-second clips at 1360x752 from text or image prompts, drawing early comparisons to OpenAI's unreleased Sora.
- Models & capabilities
- Minor
Luma AI, a startup previously known for 3D-capture (“NeRF”) tools, released Dream Machine, a model that generated short video clips from text prompts or a single still image. At launch it produced five-second clips at 1360×752 resolution, with a built-in language model that expanded terse prompts before generation.
The release came free to try, with a daily quota of ten videos and paid tiers for higher volume — a more open access model than OpenAI’s Sora, which had been previewed in February 2024 but not yet released to the public. Reviewers judged Dream Machine’s motion realism favourably against competing text-to-video systems, while noting persistent weaknesses: it struggled to render legible text and could distort fast or complex motion.
As with other video generators launched that year, Luma disclosed nothing about the composition of its training data, and journalists covering the release flagged this opacity as a limit on both scrutiny of the model’s outputs and its usefulness for professional work where provenance mattered.
Dream Machine was one of several text-to-video systems — alongside Runway’s Gen-3, Kling and, later, Sora’s public release — that arrived within months of each other in 2024, turning short AI-generated video clips from a research demo into a consumer product category.