June 13, 3D modeling platformLuma AILaunched a product calledDream MachineNew Vincent video model, available to all usersfreeOpen. Dream Machine not only supports text-generated videos, but also uses pictures as a guide to create video content. The videos it generates are comparable to OpenAI's Sora in terms of quality, motion consistency, color, light and shadow, saturation, and camera movement.
One of the technical highlights of Dream Machine is its support for physical simulation, which enables the generated videos to more realistically reflect the characteristics of the physical world, such as gravity, collisions, and changes in light and shadow. Physical simulation is particularly important in the field of cultural videos, as it ensures the coherence and consistency of the video in time and space.
Dynamic simulation and physical consistency are at the core of generating high-quality videos. This requires the model to accurately simulate the motion of objects and scenes, as well as the interactions between objects, such as gravity, collisions, friction, and air resistance. To achieve this, the video model is usually learned using a large amount of real-world video data.
The video examples generated by Dream Machine show its excellent effects. Although the efficiency of the generated video is affected by the large number of users due to its free and open use, the quality of the generated video is still very high. For example, a user generated a video of a ruins scene, which is comparable to the game scene of Unreal Engine 5; another user generated a video of a sports car running on the highway, and the camera movement and scene switching were very coherent and smooth.
Compared with OpenAI's Sora, Dream Machine and the "Keling" and other products have been fully opened for use, and their video quality is not inferior to Sora. This indicates that the competition in the Vincent video track will become more intense.
Dream Machine click experience:https://www.1ai.net/13078.html
Keling click experience:https://www.1ai.net/12558.html