Luma AI, a US-based startup focused on visual AI, has unveiled a new video generator tool called Dream Machine. This tool is designed to create high-quality, realistic videos from natural language prompts using AI technology, similar to OpenAI’s Sora and the recently released Kling model by Chinese company Kuaishou Technology.
Unlike its competitors, Dream Machine offers public access, allowing anyone to experiment with the tool. According to Luma AI, Dream Machine is a highly scalable and efficient transformer model trained directly on videos, enabling it to generate physically accurate shots. The company claims that Dream Machine is the first step towards building a universal imagination engine.
Dream Machine boasts several impressive features, such as generating 120 frames in 120 seconds and creating five-second shots with smooth motion, cinematography, and drama. Additionally, the tool can understand how humans, animals, and objects interact with the physical world, ensuring accurate physics and character consistency in the generated videos.
However, Dream Machine currently has limitations, including morphing, movements, text, and Janus, as listed on the official website. While testing the tool with the prompt “Peter Pan flying on a carpet between galaxies,” the output was described as bizarre, with Peter Pan wearing a dress, fingers warping, and no carpet present, despite the hour-long rendering time.
As Luma AI’s Dream Machine enters the market, it joins OpenAI’s Sora and Kuaishou’s Kling in the race to develop advanced video generation capabilities from natural language prompts, offering a new public option for exploring this cutting-edge technology.
