Recently, the University of Hong Kong officially launched the Goku video generation model, developed in collaboration with ByteDance. This model utilizes advanced generative algorithms to create high-quality video content based on text prompts, greatly enriching the forms of digital art.
To showcase the powerful capabilities of the Goku model, the research team created a series of impressive video examples that not only demonstrate the technical abilities of the model but also reveal its limitless potential in creative expression.
The Goku model is characterized by its efficient generation speed and image quality. By training on a vast amount of data, Goku can generate a variety of scenes, including animations, natural landscapes, and animal behaviors. The researchers tested the model using the original MovieGenBench prompts to ensure consistency and fairness in the demonstration results.
For example, one of the videos features a fashionable woman confidently strolling through the streets of Tokyo, where the colorful neon lights reflect the warm night atmosphere, and the bustling crowd creates a vivid and realistic scene.
Another video showcases several giant mammoths leisurely walking through the snow, with the surrounding snow-capped mountains and forests creating an immersive winter wonderland. These vivid scenes not only capture the audience's attention but also provide rich inspiration for artists.
Even more astonishing is that Goku also supports the direct generation of virtual digital human videos. Goku+ transforms text into hyper-realistic human videos, significantly outperforming existing methods. Notably, it can generate videos longer than 20 seconds, featuring stable hand movements and highly expressive facial and body actions of human subjects.
Additionally, it supports generating interactive videos from product images, maintaining product styles, and creating product showcase videos, as well as generating advertisement videos from text prompts.
Project link: https://saiyan-world.github.io/goku/
Highlights:
🌟 The Goku model is developed in collaboration between the University of Hong Kong and ByteDance, capable of generating high-quality video content based on text prompts.
🎨 The model showcases various scenes, including a fashionable woman strolling through Tokyo and giant mammoths walking in the snow, with vivid and realistic effects.
💡 The release of the Goku model provides new tools for visual art creation, empowering creators to explore more possibilities.
