Lumiere

Lumiere is a video-generation research project presented by Google Research and collaborating academic researchers. Its paper was first submitted in January 2024 and revised in February 2024. The official website is a research showcase with examples and a paper link. The reviewed pages did not establish a public generation app, downloadable model weights, an API or a paid service. This listing should therefore be read as a research reference, not a promise of immediate access to a video-making product.

The showcase separates several demonstrated tasks. Text-to-video examples start from written scene descriptions, while image-to-video examples animate a supplied image. Other sections illustrate style-guided generation, changing the appearance of existing footage, animating only a selected image region and filling masked parts of a video. These are author-provided demonstrations; no video was generated or independently compared during this review.

The paper's central idea is a Space-Time U-Net, an architecture that models a clip's temporal duration together instead of constructing distant keyframes and subsequently filling their gaps. This makes the project relevant to readers studying motion consistency and video diffusion methods. Its reported performance belongs to that research setting and publication date; it is not evidence that Lumiere currently outperforms available commercial tools.

No current pricing, service-level commitment or customer-data retention policy was verified. The absence of a checkout or subscription on the showcase does not establish a free product. Commercial generation rights and deployment requirements also remain unverified. Media credits do not establish rights to model outputs or third-party images. Use the official paper and examples to understand the research, and confirm access and applicable rights separately before planning a production workflow.