American start-up LumaLabs has released Dream Machine, a tool to generate short videos with a text prompt. OpenAI had previously announced Sora, but has not yet made it available.
The tool also allows users to upload a single image to integrate into the video, the site shows. Dream Machine is trained on videos, so the generated videos must look realistic and movements must physically match the space and the laws of physics.
The tool is online, but currently requires a queue of several minutes after entering a text prompt. That used to be several hours, but that seems to have been resolved. Not every prompt works; for example, a short test by Tweakers returned an error message with the first two videos. The company used AWS servers with Nvidia H100 GPUs to train the model. Many technical details are not known.
The tool can be used after logging in with a Google account. Lumalabs wants to sell subscriptions to make money.