RunwayML, a pioneer in training AI for video generation, has made its latest model, Gen-3 Alpha, available to the public. This model allows users to create hyper-realistic AI videos using text, image, or video prompts. While the previous models, Gen-1 and Gen-2, were free, Gen-3 Alpha requires users to upgrade to a paid plan starting at $12/month per editor. Despite the cost, this new model offers significant improvements in terms of speed, fidelity, consistency, and motion.
After a period of silence, RunwayML reemerged with Gen-3 Alpha as the competition in the AI video industry intensified. This model was trained on videos and images annotated with descriptive captions, enabling users to produce video clips with imaginative transitions, precise key-framing, and expressive human characters. The samples shared by Runway demonstrate the substantial upgrade that Gen-3 Alpha represents over its predecessors.
With the wide availability of Gen-3 Alpha, users can now upgrade to Runway’s paid plan and begin creating videos for various creative purposes. Initially, the model will focus on text-to-video generation, allowing users to bring their ideas to life through natural language prompts. However, in the future, Runway plans to expand its capabilities to include image-to-video and video-to-video modes. The model will also be compatible with features on the Runway platform like Motion Brush, Advanced Camera Controls, and Director Mode.
It’s worth noting that each video generated with Gen-3 Alpha will be limited to 10 seconds, with generation speed varying based on video length. While this falls short of OpenAI’s promised one-minute-long generations with Sora, it still outperforms most AI video models currently available. The creative community is eagerly anticipating how Gen-3 Alpha will be utilized. Emad Mostaque, former CEO of Stability AI, has already tested the model and compared its results to those of Sora.
Gen-3 Alpha is just the beginning for Runway, as the company plans to continuously improve the model in the coming months. They also intend to release a free version of the model in the future. Runway sees Gen-3 Alpha as the first in a series of models that will be trained on a new infrastructure designed for large-scale multimodal training. Ultimately, their goal is to build General World Models that can represent and simulate a wide range of real-world situations and interactions.