Sunday, 31 August 2025
30.3 C
Singapore
30.9 C
Thailand
29.1 C
Indonesia
27.8 C
Philippines

New AI model developed for high-resolution video generation

A Chinese research team has developed an open-source AI model, Pyramid Flow, for cost-effective, high-resolution video generation at 768p.

Researchers from Peking University, Kuaishou Technology, and Beijing University of Posts and Telecommunications have made significant progress in AI video generation. Their new AI model, Pyramid Flow, promises to revolutionise the way high-resolution virtual videos are created.

Unlike many proprietary models that require expensive resources and are often difficult to access, the team behind Pyramid Flow has chosen to make their model open-source. This move allows developers and users worldwide to access the technology freely, allowing a broader audience to experiment with and use it for various purposes.

Pyramid Flow’s cost-effective approach to high-resolution video generation

Pyramid Flow takes an innovative approach by generating videos through low-resolution stages before reaching the final high-resolution output. This multi-stage process helps to significantly reduce the computing power needed to run the model, making it more affordable and practical for users. The team claims that Pyramid Flow can produce a five-second video clip at 384p resolution in just 56 seconds, demonstrating the efficiency of their model.

One of Pyramid Flow’s most notable advantages is its ability to create high-quality, detailed imagery. The model has been shown to generate lifelike visuals, including complex scenes like underwater explosions that produce bubbles and splashing water. This level of realism is an exciting breakthrough for the AI video generation community, especially given its low cost.

Open-source availability and potential concerns

Along with the model, the team has made the source code available under the MIT License. This means that anyone can download, modify, and use the software for personal and commercial purposes without worrying about licensing fees or restrictions. The team has also provided several sample videos showcasing the impressive output quality of the model.

Additionally, the research team has made the datasets used to train Pyramid Flow available to the public. These datasets consist of approximately 10 million short videos, allowing other developers to build upon and improve the model in the future.

However, using open-source datasets in AI video generation has raised some concerns. Critics argue that such practices could infringe on the intellectual property rights of copyright holders. While the team behind Pyramid Flow has yet to address these concerns directly, they have suggested that their model could be a valuable tool for fine-tuning open-source material. This would help reduce reliance on third-party sources, alleviating some copyright concerns.

Pyramid Flow represents a significant leap forward in AI video generation technology. It offers both high-quality output and an open-source approach that could open up new possibilities for developers and creators. The cost-effective nature of the model and the free access to the underlying code and datasets could reshape the way AI-generated videos are used across industries, making high-resolution video creation more accessible than ever.

Hot this week

Xbox introduces cross-device play history for seamless gaming

Xbox rolls out cross-device play history, syncing recently played and cloud-enabled games across consoles, PC, and handheld devices.

Google rolls out QR code verification for secure messaging

Google is testing QR code verification in Messages, making it easier for users to confirm the identity of contacts and secure RCS chats.

ASUS ROG unveils new OLED gaming monitors with tandem technology at Gamescom 2025

ASUS ROG introduces new OLED gaming monitors at Gamescom 2025, featuring Tandem OLED technology, higher brightness, and longer lifespan.

HPE introduces agentic AI innovations for self-driving network operations

HPE enhances its Juniper Mist platform with new agentic AI features, bringing self-driving capabilities to network operations.

Meta partners with Midjourney to bring AI-generated images to its platforms

Meta partners with Midjourney to bring advanced AI-generated images to its platforms, boosting creative features across its apps.

Meta introduces new AI safeguards to protect teens from harmful conversations

Meta is strengthening AI safeguards to prevent teens from discussing self-harm and other sensitive topics with chatbots on Instagram and Facebook.

ChatGPT to introduce parental controls as AI safety concerns rise

OpenAI is introducing parental controls for ChatGPT, addressing growing concerns about the safety of AI chatbots and their impact on young users.

Japan uses an AI simulation of Mount Fuji’s eruption to prepare citizens

Japan uses AI to simulate a Mount Fuji eruption, showing its potential devastation and promoting disaster preparedness.

Anthropic updates Claude chatbot policy to use chat data for AI training

Anthropic will utilise Claude chatbot conversations for AI training starting from 28 September, with opt-out options and a five-year data retention policy.

Related Articles

Popular Categories