Skip to content
Annuaire
Sections
Annuaire des entreprises

Luma AI, from 3D capture to generative video

L’essentiel

Based in the United States, Luma AI develops artificial intelligence models capable of producing visual content, particularly videos from text or images. Known for its early 3D capture tools, the company now caters to creators and professionals with its Dream Machine platform.

À retenir

Based in the United States, Luma AI develops artificial intelligence models capable of producing visual content, particularly videos from text or images. Known for its early 3D capture tools, the company now caters to creators and professionals with its Dream Machine platform.


Luma AI is an American artificial intelligence company based in Palo Alto, California. Its focus is the generation and representation of visual environments, at the intersection of computer vision, 3D and video. Available at lumalabs.ai, its offering aims to reduce the technical constraints of creating moving images. It targets both individual experimentation and professional uses, such as planning sequences, visual communication and exploring advertising concepts.

From real scenes to generated worlds

Founded in 2021 by Amit Jain, Alex Yu and others, Luma AI first became known for tools that could reconstruct three-dimensional objects and scenes from captured images. This approach drew on advances in neural scene representations, including neural radiance fields, or NeRFs. The aim was to make 3D capture more accessible, particularly from a smartphone, without always relying on specialized equipment.

The company then expanded its scope to content generation. With Genie, it offered the creation of 3D objects from text descriptions. The launch of Dream Machine in June 2024 marked a new milestone: producing videos from written instructions or images supplied by the user.

Tools for creating and exploring sequences

Dream Machine is the main entry point to its generative video offering. The user describes a scene or submits an image, and the model then produces an animated sequence. Instructions can specify the subject, visual atmosphere or desired movement. This approach makes it possible to test several directions without immediately organizing a shoot or building every element in animation software.

Unveiled in January 2025, the Ray2 model continues in this direction, with announced efforts to improve the consistency of movement and physical interactions. These aspects are crucial for generative video: a convincing result must preserve the identity of subjects and the continuity of the scene across frames. Outputs may nevertheless contain inconsistencies and require selection or even retouching.

Luma AI sells access to its tools through various plans and also offers an API to integrate its capabilities into third-party applications. Its positioning thus combines a ready-to-use creation service with a technical building block for developers.

What next?

For Luma AI, the challenge is to turn sequence generation into a process that is controllable enough to fit into production pipelines. The precision of instructions, character consistency, editing options and usage costs will matter as much as visual quality. In a market that includes OpenAI, Google and Runway, differentiation will also depend on the user experience. Questions around content rights, the provenance of training data and the identification of synthetic media will remain central to earning the trust of professional clients.

Sur votre appareil

Comprendre cet article

L’analyse utilise l’intelligence locale du navigateur lorsqu’elle existe, sinon un résumé extractif. Le texte n’est envoyé à aucun service extérieur.

Facebook X LinkedIn

Ensuite A lire aussi

Annuaire des entreprises

Sentry, error tracking for developers

US-based Sentry helps software teams identify errors and slowdowns affecting their applications. Born out of an open source…

3 min read

Free, no spam, one-click unsubscribe.