Luma Dream Machine: New AI model dreams up cinematic videos

Luma AI promises high-quality 5-second videos and invites you to take a free test. It shows: The AI model is not free of errors, but it is still impressive.

Save to Pocket listen Print view

(Image: Olya Detry/Shutterstock.com)

4 min. read
By
  • Jan Mahn
This article was originally published in German and has been automatically translated.

It takes two minutes for the AI model called "Luma Dream Machine " to generate a 5-second film snippet from a prompt or image template, 120 frames in 120 seconds. The AI comes from a start-up called Luma AI Inc. and has quickly triggered a new trend in social networks: turning well-known still images, especially memes, into short videos. This is possible because Luma has released its model for free testing and has not yet implemented any restrictions on the number of videos. Due to the high demand, videos sometimes end up in the queue for a few minutes.

Empfohlener redaktioneller Inhalt

Mit Ihrer Zustimmmung wird hier ein externes Video (Kaltura Inc.) geladen.

Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Kaltura Inc.) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.

To try it out, all you need is a Google account to log in; the operator of the AI can see the email address, name and profile picture of this account via this login. You can start immediately after logging in: You type a prompt into the input field and can optionally upload a picture. We made our first attempts without an uploaded image and obtained results reminiscent of the early days of the Midjourney and Dall-E image generators. Cinematic camera movements were achieved very well by the AI straight away, but objects did not look like they would in reality. An attempt to generate a horror movie clip showing a man with a chainsaw resulted in a creative mix of chainsaw and hedge trimmer.

Does not correspond to reality: Luma dreams up a hedge saw, a mixture of chainsaw and hedge trimmer.

The next test: a robot that loads a dishwasher. This dream of everyone who has no desire to do housework will initially remain a dream – Luma dreamed up a world in which robotic arms can reach through dishes. However, there is nothing wrong with the camera work or the lighting. The third test then confirmed the thesis: Luma is better the closer the wishes in the prompts are to what has already been photographed or filmed, i.e. what was most likely in the training data. The AI was able to generate a simple robot vacuum cleaner driving through the kitchen without any problems.

That was nothing: a robot that puts away a dishwasher, that was the wish in the prompt. Luma created a parallel world in which the laws of physics do not apply and robots reach through objects.

The videos become really good when you upload photos and briefly describe what you can see. In this way, Luma adds an impressive temporal dimension to the images and dreams the story of the still image a few seconds further – almost free of artifacts that still dominate when operating without an image template. For the test, we used lead photos from the c't section "Vorsicht, Kunde!", among others, in which customer problems with companies are symbolically re-enacted. The 5-second videos that bring the still images to life are deceptively real and error-free.

Empfohlener redaktioneller Inhalt

Mit Ihrer Zustimmmung wird hier ein externes Video (Kaltura Inc.) geladen.

Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Kaltura Inc.) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.

In recent days, the new art form of turning well-known memes – such as the girl grinning in front of the burning house – into videos with Luma has emerged on the X platform. Others animated photos of (deceased) relatives or historical photographs.

Behind the company Luma AI Inc. are young founders such as Alex Yu, who studied at the University of California Berkeley and then completed internships at Google and Adobe. The list of model developers also includes Dominic Rampas from Germany, who studied in Ingolstadt and was already involved with AI during his studies. Luma AI does not yet have a recognizable business model - because the model is already delivering good results in the first test phase, there is a high probability that one of the major AI companies will soon approach the founders with an interest in buying it.

(jam)