Introducing Runway’s Gen-2 AI Video Model
Runway, the startup behind the AI art generator Stable Diffusion, is preparing to launch the first public test for its Gen-2 AI video model. This new model claims to be the first publicly available text-to-video model and is a significant upgrade from the company’s Gen-1 video-to-video AI model. Gen-2 aims to allow users to create 3-second videos from scratch using simple text prompts.
Google and Meta Entering the Text-to-Image Race
Both Google and Meta are working on their text-to-image generators, but have been tight-lipped about their progress. In contrast, Runway has been steadily growing its reputation with its video editing tools and AI models, starting with its Gen-1 AI model that could transform existing videos based on text prompts or reference images.
ModelScope – An Alternative AI Video Generator
For those eager to try AI video generation, ModelScope, an AI text-to-video system developed by the DAMO Vision Intelligence Lab (a research division of Alibaba), is another option. ModelScope has generated buzz for its sometimes awkward and quirky 2-second video clips created using a basic diffusion model.
Open Source and GPU Server Requirements
ModelScope is open source and available on Hugging Face, but users may need to pay a small fee to run the system on a separate GPU server. Tech YouTuber Matt Wolfe offers a tutorial on setting it up, or users with technical skills and sufficient VRAM can run the code themselves.
Copyright Concerns with Training Data
ModelScope’s training data seems to include a significant number of videos and images from Shutterstock, as many generated videos show the stock photo site’s logo. This issue is not unique to ModelScope; Getty Images has sued Stability AI for similar concerns with its AI art generator, Stable Diffusion.
Runway Aiming to Stand Out in AI Research
Runway seeks to make a name for itself in the competitive AI research field. Their Gen-1 system relies on a large-scale dataset of images, videos, and text-image data, but the researchers have found a lack of high-quality video-text datasets. It remains to be seen how Runway’s more polished text-to-video model will compare to heavyweights like Google in the future.
More content at UsAndAI. Join our community and follow us on our Facebook page, Facebook Group, and Twitter.