The New York start-up specializing in artificial intelligence has just unveiled Gen-3 Alpha. The first in a series of models trained on new infrastructure designed for large-scale multimodal training, this model represents a major improvement in fidelity, consistency and motion over Gen-2. For the record, the latter was released in June 2023, just a few months after Gen-1, published in February.
Trained jointly on videos and images from public and private datasets, Gen-3 Alpha must power a certain number of tools from Runway. This includes Text to Video, Image to Video and Text to Image tools, existing control modes such as Motion Brush, Advanced Camera Controls, Director Mode as well as upcoming tools for fine-tuned control of structure, style and movement, the company says. The generated video clips can range from 5 to 10 seconds.
Runway ready to overshadow OpenAI’s Sora
With this model, Runway paves the way for further advances in video generation from different media. Gen-3 Alpha is capable of generating expressive human characters with a wide range of actions, gestures and emotions, which opens up new storytelling possibilities, especially for use in commercials or short films, e.g. example.
So far, the only comparable tool is OpenAI’s Sora, unveiled last February and capable of creating one-minute scenes with 1080p resolution. Obviously, other technology giants have published text-to-video tools, like Google with Imagen Video and Veo or Meta with Make-A-Video. However, the multimodal approach chosen by Runway seems more convincing at first glance.
Custom versions coming soon
Researchers, engineers and artists worked together on the development of this model, we learn. “It was designed to interpret a wide range of cinematic styles and terminologies.” Runway therefore offers to create personalized versions of Gen-3 within the framework of partnerships established with the main entertainment and media organizations.
The company says Gen-3 Alpha will be released with a set of safeguards, including its improved internal visual moderation system and C2PA provenance standards. For the moment, no publication date has been given, so we will have to make do with the demonstration videos.
Selected for you
