Follow zdnet: Please add it as a priority source On Google.
Important points of ZDNET
- According to Luma AI, Ray3 has multimodal reasoning capabilities.
- AI developers are pushing video models into the creative industry.
- Some tools are positioned as automated creative partners.
Just a few years ago, video clips generated by AI were laughing stocks on the internet – everyone remembers nightmare videos generated by AI Will Smith blows spaghetti? Since then, this technology has come a long way. Today, tech startups are competing to provide generative AI tools that rival the quality of Hollywood production studios, at least in their vision of the future.
Also: This new AI video editor is an all-in-one production service for filmmakers – how to try it
Its latest developments in the competition include AI startups Luma ai On Thursday, the new video generation model Ray3 was announced. Another product, Luma Dream Machine, allows users to create videos from just photos.
Models are currently available Dream Machine. You can also pay to Adobe's Firefly and Creative Cloud Pro customers.
“Inference” AI video model
I've heard of so-called inference models like Openai's O3. This is thought to consider more time-consuming queries, especially complex, than standard-generated AI models, to return a more informative and thorough answer. However, so far these models have not had video generation capabilities.
Also: AI is not “inference” at all – this team exposes industry hype
According to Luma AI, Ray3 is distinguished from competing tools from companies like Google, Runway, Meta and Openai.
This is a vaguely defined, ontologically controversial term that has been sold a lot these days, just as AI systems are a bit like “understanding”, “creativity”, and “agency.” Simply put, it refers to the ability of a model to break down a problem into multiple steps, reflect the quality of its output, and repeatedly improve it over time.
Rather than generating videos from text prompts, Ray3 divides the production process into multiple steps, just like creative teams do. It has a multimodal inference function. This means that you can generate text along with visual assets that help you sketch the concept of the final video to the user.
For example, filmmakers can encourage the system to annotate images or suggest camera angles for a series of shots. According to Luma AI, the model is also the first of its kind to enable it to provide video output over a high dynamic range of 4K. That is, it provides a wider range of visual light and shadows.
Also: Does AI damage human creativity? Most Americans say yes
“The results are characters that appear to be consistent, scenes that unfold naturally over time, and videos that feel more consistent with each behavior,” Luma Ai wrote press release.
Luma AI has not published any restrictions on the length of video that Ray3 can generate, and the company did not immediately respond to ZDNET's request for comments on the subject.
Automated Creative Partners
Luma AI has Ray3 as an automated creative partner for filmmakers, video game designers and advertisers.
For example, the new “Draft Mode” allows Ray3 to quickly generate a variety of test clips. Each has subtle variations, offering creative teams a variety of options, saving time in the idea process.
“This allows creators to enter a state of flow and experiment freely without worrying about calculating time or cost,” writes Luma Ai.
Also: AI models know when they are being tested – and change their behavior, research shows
Other AI developers sell their own tools on the premise that they can serve as automated creative partners, allowing humans to offload time-intensive and everyday tasks, saving money in the process. For example, earlier this week, Amazon announced an AI agent that can help brands with almost every step of the process of creating short video ads.
