The thin line between the real and artificial realms was blurred further the other day when the US-based startup Luma AI introduced a new video generator, Dream Machine, similar to OpenAI’s Sora. Luma AI claims that the new tool is a next-generation video model capable of creating high-quality, realistic shots from prompts in natural language using AI.
“Dream Machine is an AI model that makes high quality, realistic videos fast from text and images,” reads the introductory text on the official website. Dream Machine comes close on the heels of Kuaishou Technology, a Chinese company known for its short-video platform, launching its text-to-video model named Kling. The difference is that unlike Sora and Kling, Dream Machine offers access to the public.
Luma AI describes Dream Machine as a highly scalable and efficient transformer model trained directly on videos which makes it capable of generating physically accurate shots. The company claims that Dream Machine, its first step towards building a universal imagination engine, generates 120 frames in 120 seconds and it can iterate faster. The tool reportedly generates five second shots with smooth motion, cinematography, and drama. Dream Machine can also understand how humans, animals, and objects interact with the physical world allowing it to create videos with great character consistency and accurate physics and helps in experimentation with an endless array of fluid, cinematic, and naturalistic camera motions.
Flashback to November 2022 when the generative AI revolution was ushered in with Open AI’s ChatGPT making its debut, prompting some some startlingly human-like artistic creations – with tools including Suno, an AI-powered song generator. Suno can create lyrics to vocals and instrumentation, from a simple prompt and that too steered towards a precise genre. While the free version of Suno has limitations, a subscription-based plan allows to use the songs commercially, for example on YouTube or even uploading them to Spotify or Apple Music. As Suno’s FAQ section says: “the availability and scope of copyright protection for content generated (in whole or in part) using artificial intelligence is a complex and dynamic area of law, which is rapidly evolving and varies among countries”.
With tools like Dream Machine and Suno, video and audio content generation is entering the next level, with immense commercial potential and the possibility of drastically reducing the time and cost associated with production. However, the industry will need to address the ethical and legal challenges surrounding the use of training data and the potential for misuse in creating deepfakes or spreading misinformation. Despite these concerns, the allure of sophisticated, near-instant video and audio generation is likely to attract many creators and companies looking to push the boundaries of what’s possible with AI-powered tools.
As the technology continues to mature, it will be crucial for the companies concerned to strike a balance between innovation, accessibility and responsibility. There’s also a wider debate around AI-generated content looming in the background. The New York Times is suing OpenAI and Microsoft because it claims ChatGPT was trained on millions of its articles without its permission. Is training an AI model on someone else’s content infringing on its copyright? That’s the big unanswered question.
The viral ‘Heart on my sleeve’ in May 2023, which was supposedly made by Drake and The Weeknd and racked up more than 11mn views on various platforms, before it was revealed that it’d been made using AI by a user called Ghostwriter977. This resulted in a take-down notice from the artists’ record label, Universal Music Group, and a copyright debate that’s still rumbling on. But attempting to crack down on AI-generated music and video may pose a unique challenge. The legal landscape for AI work remains unclear, the tools to create it are widely accessible and social media makes it easier than ever to distribute it.