U.S. company and universities create new technology to generate animated videos
Categoria
🎥
CurtaTranscrição
00:00Researchers from NVIDIA and universities such as Stanford and Berkeley have developed an AI model called TTTMLP, Test Time Training Multilayer Perceptron, that generates animated videos of up to one minute in the Tom and Jerry style, based on text, storyboards, prompts.
00:18Using test time training layers, the model enables the creation of complex narratives with dynamic scenes and continuity. For example, Tom taking an elevator in New York, sitting at a desk, and working on a computer.
00:31But everything changes when Jerry appears and cuts the computer's power cord. What follows is a classic cat and mouse chase through the office.
00:40Despite the promising results, the videos still show some visual flaws. Scientists believe the technique could evolve to produce longer, more sophisticated stories with improved output.