AI news
June 13, 2024

Dream Machine is a new AI video generator you can try today

Luma Labs just released a new AI image generator that challenges Sora and Kling.

Jim Clyde Monge
by 
Jim Clyde Monge

Less than a week after the Chinese startup Kuaishou announced its impressive AI video generator Kling to compete with OpenAI’s highly anticipated video generator Sora, a new competitor has emerged.

Luma Labs has launched a new AI video generator model called Dream Machine.

Unlike other AI tools, you can try Dream Machine today with no waitlist whatsoever.

What is “Dream Machine”?

Luma Labs’ Dream Machine is a next-generation AI video generator that can create high-quality, realistic videos based on text prompts and images.

Key features:

  1. Fast Generation: Dream Machine is an incredibly fast video generator! 120 frames in 120s.
  2. Realistic and Consistent Motion: The advanced modeling ensures the videos are both highly realistic and seamlessly consistent.
  3. High Character Consistency: Maintain character integrity while simulating real-world physical characteristics, ensuring your videos are coherent and lifelike.
  4. Natural Camera Movements: Smooth, natural camera movements that perfectly match the emotional tone of each scene.

These features are indeed impressive, but we need to test them out and see for ourselves.

Text-to-video Example

Head over to Luma Lab’s Dream Machine page and click on the “Try Now” button in the top right corner.

Head over to Luma Lab’s Dream Machine page and click on the “Try Now” button in the top right corner.
Image by Jim Clyde Monge

Create your account, and you’ll see a text box where you describe the video you want to generate.

Create your account and you’ll see a text box where you describe the video you want to generate. Dream Machine luma
Image by Jim Clyde Monge

Note: Due to an extremely high demand, it’ll take a few minutes for your video to be generated.

Here’s my first video generated with Dream Machine:

Prompt: A teddy bear in sunglasses playing electric guitar, dancing and headbanging in the jungle in front of a large beautiful waterfall
Prompt: A teddy bear in sunglasses playing electric guitar, dancing and headbanging in the jungle in front of a large beautiful waterfall
Video by Jim Clyde Monge

You can view the full video from this direct link.

Unlike other competitors like Stable Video, Runway, and Pika, which simply pan and zoom 2D images, Dream Machine actually does scene transitions and moves subjects realistically.

Here’s another example:

Prompt: Low-angle shot of a majestic tiger prowling through a snowy landscape, leaving paw prints on the white blanket

The result is coherent with the prompt, and the quality of the video looks good. However, the tiger left no paw marks on the snow, which is a bit odd.

Image-to-video Example

Now, let’s try this static image generated with Midjourney V6.

Prompt: Couple on gravel giant road bikes. Open road. Midday in 2025 Greece Desert landscape, dry river in the background, distant mountains, a diner sign. Wide shot, full body, side angle. In the style of wes anderson, Bright sunlight, dust trail, saturated colors.
Prompt: Couple on gravel giant road bikes. Open road. Midday in 2025 Greece Desert landscape, dry river in the background, distant mountains, a diner sign. Wide shot, full body, side angle. In the style of wes anderson, Bright sunlight, dust trail, saturated colors.
Image by Jim Clyde Monge

You can drag and drop the image and add the same prompt on the Dream Machine.

Image by Jim Clyde Monge

Here’s the final result:

Video by Jim Clyde Monge

Here’s one generated by X user Tatiana Tsiguleva.

Reference image generated with Midjourney:

image generated with midjourney, a girl with birds at the back
Image by X user @ciguleva

Video generated by Dream Machine:

an animated video of a girl with birds at the back
Video by X user @ciguleva

Watch it in video format here.

This looks incredible. The VFX guys are probably sweating to see these kinds of sequences generated by AI in less than 2 minutes.

Safety and Ethical Considerations

As with any AI tool, it’s crucial to consider the safety and ethical implications of using Dream Machine. AI-generated videos can be incredibly lifelike, which opens the door to potential misuse.

Luma Labs appears to be aware of these concerns.

I tried the prompt below to test how their AI model handles such inputs.

Prompt: A gory 1980s war scene where soldiers are firing at each
Dream machine failed to generate
Image by Jim Clyde Monge

Dream Machine rejected the prompt and showed an error “Failed to create generation for requested prompt.” Kudos to Luma for implementing these safeguards.

However, there will always be users who try to trick the model and find ways to generate unsafe content. We’ll see in the coming days how effective these measures are and if additional precautions are needed.

Is it better than Sora or Kling?

The best way to judge is to compare the results side by side. Here’s a video of a subject eating a hamburger generated by Dream Machine, Kling, and Sora.

An video of a boy wearing eyeglasses and eating a hamburger
The left video was generated with Sora. The right video generated with Kling

Looking at these videos, Sora and Kling are still superior to Dream Machine. But hey, the latter videos are carefully hand-picked examples, who knows how many attempts it took to achieve such quality?

How much does it cost?

Before you get too excited about the free access to Dream Machine, it’s important to note that the service operates on a freemium model. This means you only get 30 free generations per month, and you must pay for extra credits if you want to generate more AI videos.

Here’s the pricing information:

Dream Machine pricing
  • Standard ($29.99): 120 generations per month
  • Pro ($99.99): 400 generations per month
  • Premier ($499.99): 2,000 generations per month

Final Thoughts

Overall, it’s incredibly generous for Luma to allow the public access to its AI video generator. Although 30 free generations per month might seem limited, it’s enough to test and judge the video quality and coherence of the results.

Now, let’s talk about the quality. Is it on par with OpenAI’s Sora? Close, but not quite. Dream Machine still struggles with temporal coherence. However, it’s definitely better than Runway or Pika.

Luma is aware of Dream Machine’s limitations. They even mentioned them on their website.

Dream Machine current limitations

The initial user feedback and user-generated videos should provide valuable data to further train and tune the model for better results in the next version.

Get your brand or product featured on Jim Monge's audience