Open source AI video generator Pyramid Flow now available online

Already becoming popular in YouTube tutorial videos, Pyramid Flow is a new AI model that was trained on open-source datasets, which total around 10 million videos. The work of a team that brings together AI researchers from Peking University, Kuaishou Technology, and Beijing University of Posts and Telecommunications, Pyramid Flow, is itself open-source. Covered by the terms of the MIT License, it can generate virtual hi-resolution (768p) video content, and it obviously shines when having to deal with 384p. According to its creators, it can come up with a five-second video in less than a minute (using the A100 GPU in an undisclosed hardware configuration, apparently).

In some scenarios, Pyramid Flow excels. When dealing with certain text prompts, however, the results are far from satisfactory. As it still is the case with many generative AI applications, working with such a tool comes with a certain amount of randomness. On the bright side, the approach of this model uses far less computing power than its competitors, and, since its code is open-source, it can be used in local or cloud applications by those interested without having to bother with the licensing part.

However, the AI researchers who created Pyramid Flow listed all the datasets used in training it but failed to mention anything about the copyright issues that may arise. While some content creators claim that virtual videos made using open-source material violate the rights of the copyright holders, Pyramid Flow could be useful in fine-tuning this kind of content without having to involve any third parties.

Those interested in AI from an open-source perspective can grab Open-Source AI: The Future of Data Research by J.D. TjonWolf off Amazon for $9.99 (Kindle) or $12.99 (paperback).