Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
The number of AI video generation models continues to grow with a new one, Pyramid Flow, launching this week and offering high quality video clips up to 10 seconds in length — quickly, and all open source.
Developed by a collaboration of researchers from Peking University, Beijing University of Posts and Telecommunications, and Kuaishou Technology — the latter the creator of the well-reviewed proprietary Kling AI video generator — Pyramid Flow leverages a new technique wherein a single AI model generates video in stages, most of them low resolution, saving only a full-res version for the end of its generation process.
It’s available as raw code for download on Hugging Face and Github, and can be run in an inference shell here but requires the user to download and run the model code on their own machine.
Advertisement
NEW: Open Source Text/ Image to video model is out – MIT licensed – Rivals Gen-3, Pika & Kling 🔥
At inference, the model can generate a 5-second, 384p video in just 56 seconds—on par with or faster than many full-sequence diffusion counterparts — though Runway’s Gen 3-Alpha Turbo still takes cake in terms of speed of AI video generation, coming in at under one minute and often times 10-20 seconds in our tests.
We haven’t had a chance to test Pyramid Flow yet, but the videos posted by the model creators appear to be incredibly lifelike, high enough resolution, and compelling — analogous to those of proprietary offerings. You can see various examples here on its Github project page.
Indeed, Pyramid Flow is available designed now to download and use — even for commercial/enterprise purposes — and is designed to compete directly with paid proprietary offerings such as Runway’s Gen-3 Alpha, Luma’s Dream Machine, Kling, and Haulio, which can cost hundreds of even thousands of dollars a year for users on unlimited generation subscriptions.
Advertisement
As the race between various AI video providers to gain users continues, Pyramid Flow aims to bring more efficiency and flexibility to developers, artists, and creators seeking advanced video generation capabilities.
A new technique for high-quality AI videos: ‘pyramidal flow matching’
AI video generation is a computationally intensive task that typically involves modeling large spatiotemporal spaces. Traditional methods often require separate models for different stages of the process, which limits flexibility and increases the complexity of training.
Pyramid Flow is built on the concept of pyramidal flow matching, a method that drastically cuts down the computational cost of video generation while maintaining high visual quality, completing the video generation process as a series of “pyramid” stages, with only the final stage operating at full resolution.
The authors include Yang Jin, Zhicheng Sun, Ningyuan Li, Kun Xu, Hao Jiang, Nan Zhuang, Quzhe Huang, Yang Song, Yadong Mu, and Zhouchen Lin. Most of these researchers are affiliated with Peking University, while others are from Kuaishou Technology.
As they write, the ability to compress and optimize video generation at different stages leads to faster convergence during training, allowing Pyramid Flow to generate more samples per training batch.
For example, the proposed pyramidal flow reduces the token count by a factor of four compared to traditional diffusion models, which results in more efficient training.
The model can produce 5- to 10-second videos at 768p resolution and 24 frames per second, all while being trained on open-source datasets. Specifically, the paper states that Pyramid Flow was trained on trained on:
Advertisement
LAION-5B, a large dataset for multimodal AI research.
CC-12M, a dataset of web-crawled image-text pairs.
SA-1B, which features high-quality, non-blurred images.
WebVid-10M and OpenVid-1M, which are video datasets widely used for text-to-video generation.
In total, the authors curated approximately 10 million single-shot videos.
However, many of these “public” or “open source” datasets have in recent years come under fire from critics for including copyrighted material without permission or informed consent of the copyright holders, and LAION-5B in particular accused of hosting child sexual abuse material.
Separately, Runway is among the companies being sued by artists in a class action lawsuit for training on materials without permission, compensation, or consent — allegedly in violation of U.S. copyright. The case remains being argued in court, for now.
Permissively licensed, open source for commercial usage
Pyramid Flow is released under the MIT License, allowing for a wide range of uses, including commercial applications, modifications, and redistribution, provided the copyright notice is preserved.
This makes Pyramid Flow an attractive option for developers and companies looking to integrate the model into proprietary systems, and could challenge Luma AI and Runway as both look to offer paid application programming interfaces for developers seeking to integrate their proprietary AI video generation technology into customer or employee-facing apps.
Advertisement
Yet those proprietary models already exist as inferences suitable for developers, while Pyramid Flow has a demo inference on Hugging Face, it is not suitable for building full applications atop it and users would need to host their own version of an inference, which could also be costly, despite the model itself being “free.”
Using Pyramid Flow, Lionsgate or any other film studio could fine-tune the open source version without paying a third party company. However, they would still need to have on hand or contract out the developer talent and computing resources necessary to do so, which may make partnering with established AI providers such as Runway more appealing, since that company and others like it already have the AI engineering talent at their disposal in house.
The research team behind Pyramidal Flow Matching has also made a commitment to openness and accessibility. All code and model weights will be made freely available to the public through their official project page, ensuring that researchers and developers around the world can utilize and build upon this work.
Advertisement
Despite its strengths, Pyramid Flow does have some limitations. For now, it lacks some of the advanced fine-tuning capabilities found in models like Runway Gen-3 Alpha, which offers precise control over cinematic elements like camera angles, keyframes, and human gestures. Similarly, Luma’s Dream Machine provides advanced camera control options that Pyramid Flow is still catching up to.
Moreover, the relatively recent launch of Pyramid Flow means its ecosystem—while robust—isn’t as mature as those of its competitors.
Looking ahead: AI video race shows no signs of slowing
As the AI video generation market continues to evolve, Pyramid Flow’s launch signals a shift toward more accessible, open-source solutions that can compete with proprietary offerings such as Runway and Luma.
For now, it offers a solid alternative for those looking to avoid the cost and limitations of closed models, while providing impressive video quality on par with its more commercial counterparts.
Advertisement
In the coming months, developers and creators will likely keep a close eye on Pyramid Flow’s growth. With the potential for further improvements and optimizations, it could very well become a go-to tool in the arsenal of video content creators everywhere. All the companies and researchers are currently battling both for technological supremacy and users.
South Korean memory giant SK Hynix has announced it has begun the mass production of the world’s first 12-layer HBM3E, featuring a total memory capacity of 36GB, a huge increase from the previous 24GB capacity in the 8-layer configuration.
This new design was made possible by reducing the thickness of each DRAM chip by 40%, allowing more layers to be stacked while maintaining the same overall size. The company plans to start volume shipments by the end of 2024.
The HBM3E memory supports a bandwidth of 9600 MT/s, translating to an effective speed of 1.22 TB/s if used in an eight-stack configuration. The improvement makes it ideal for handling LLMs and AI workloads that require both speed and high capacity. The ability to process more data at faster rates enables AI models to run more efficiently.
Nvidia and AMD hardware
For advanced memory stacking, SK Hynix employs innovative packaging technologies, including Through Silicon Via (TSV) and the Mass Reflow Molded Underfill (MR-MUF) process. These methods are essential for maintaining the structural integrity and heat dissipation required for stable, high-performance operation in the new HBM3E. The improvements in heat dissipation performance are particularly important for maintaining reliability during intensive AI processing tasks.
In addition to its increased speed and capacity, the HBM3E is designed to offer enhanced stability, with SK Hynix’s proprietary packaging processes ensuring minimal warpage during stacking. The company’s MR-MUF technology allows for better management of internal pressure, reducing the chances of mechanical failures and ensuring long-term durability.
Advertisement
Early sampling for this 12-layer HBM3E product began in March 2024, with Nvidia’s Blackwell Ultra GPUs and AMD’s Instinct MI325X accelerators expected to be among the first to use this enhanced memory, taking advantage of up to 288GB of HBM3E to support complex AI computations. SK Hynix recently rejected a $374 million advanced payment from an unknown company to ensure it could provide Nvidia with enough HMB for its in-demand AI hardware.
“SK Hynix has once again broken through technological limits demonstrating our industry leadership in AI memory,” said Justin Kim, President (Head of AI Infra) at SK Hynix. “We will continue our position as the No.1 global AI memory provider as we steadily prepare next-generation memory products to overcome the challenges of the AI era.”
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
NASA’s Europa Clipper mission, set to visit the icy moon of Jupiter, was set to launch from the Kennedy Space Center in Florida this week but had its launch delayed because of Hurricane Milton. Now, NASA has announced that it is targeting no earlier than Monday, October 14, for the launch, and we’ve got the details on how you can watch the event live.
The mission intends to explore Europa, the moon of Jupiter that has a liquid water ocean beneath a thick, icy shell. Because of the presence of liquid water there, scientists want to learn whether the moon could be potentially habitable, as it is one of the most promising locations that life could survive outside of Earth. The mission will search for information about the ocean and the presence of the building blocks of life, called organic compounds, to see if the ingredients for life are present there.
Europa Clipper had been scheduled to launch this week on Thursday October 10, but the launch was postponed because of the hurricane conditions around the Kennedy Space Center. The spacecraft had to be secured against the high winds and heavy rain in its hanger at Launch Complex 39A.
“The safety of launch team personnel is our highest priority, and all precautions will be taken to protect the Europa Clipper spacecraft,” said Tim Dunn, senior launch director at NASA’s Launch Services Program, at the time. “Once we have the ‘all-clear’ followed by facility assessment and any recovery actions, we will determine the next launch opportunity for this NASA flagship mission.”
NASA Kennedy confirmed that it was all clear after the storm, and now teams are continuing to check the status of the spacecraft and the ground systems, but NASA has confirmed it is targeting Monday onward for the launch.
The launch will be livestreamed by NASA, which you can watch either using the NASA+ app or using the YouTube video embedded above. The YouTube video currently has a date of November 6 on it, but this is just a placeholder date and not when the actual launch will take place.
The exact time of the launch hasn’t been announced yet, but you can follow NASA’s X (formerly Twitter) account to get details as soon as the time is made public.
If you love streaming TV, Xbox games, and as little clutter as possible, then you need to pick up this Amazon Fire TV bundle deal that gets you set up with all the Xbox games you could ever want to play. The regular price for this bundle is $146.97, but Amazon currently has it on sale for $79.99. At that price, you’re getting a very good value and, it’s the lowest this bundle has ever been. Prior to this discount, which is 45% off the regular price, this bundle was $111.97. You’re basically saving another $32 by picking it up now as opposed to its previous price.
The best thing about this bundle is that you get everything you need to play Xbox games. The Fire TV Stick 4K Max gives you access to the Xbox Game Pass app. The bundle also comes with a 30-day subscription to Xbox Game Pass Ultimate. So once you get that 30-day sub applied, all you need is a controller. As luck would have it this bundle comes with an Xbox Wireless Controller as well. The Sky Cipher Special Edition model to be exact.
If you don’t like the Sky Cipher color, Amazon is offering the bundle with the Astral Purple and Deep Pink colors as well. Xbox Game Pass is great because you get access to tons of games in Microsoft’s curated list, which is always adding new titles to play. Now since this is on a Fire TV Stick 4K Max, you’ll be playing Xbox games in the cloud through the cloud feature of Xbox Game Pass Ultimate. So you will need an internet connection.
In addition to the games, the Fire TV Stick 4K Max gives you access to tons of other streaming apps. Including Disney Plus, Netflix, Hulu, Apple TV Plus, Prime Video, and many more. It’s definitely a killer deal you don’t want to pass up.
The Dell PowerEdge M1000e modular blade enclosure is a breakthrough in enterprise server architecture. Built from the ground up to combat data center sprawl and IT complexity, the M1000e delivers one of the most energy-efficient, flexible and manageable blade server products on the market. Flexible and scalable, the M1000e is designed to support future generations of blade technologies regardless of processor or chipset architecture. The M1000e is optimized for use with all Dell PowerEdge blades.
In this video, Lonnie Laub, STI’s lead technician, guides you through the features and benefits of the M1000e. http://www.stikc.com .
TikTok’s executives and employees were well aware that its features foster compulsive use of the app, as well as of its corresponding negative mental health effects, according to NPR. The broadcasting organization reviewed the unredacted documents from the lawsuit filed by the Kentucky Attorney General’s Office as published by the Kentucky Public Radio. More than a dozen states sued TikTok a few days ago, accusing it of “falsely claiming [that it’s] safe for young people.” Kentucky Attorney General Russell Coleman said the app was “specifically designed to be an addiction machine, targeting children who are still in the process of developing appropriate self-control.”
Most of the documents submitted for the lawsuits had redacted information, but Kentucky’s had faulty redactions. Apparently, TikTok’s own research found that “compulsive usage correlates with a slew of negative mental health effects like loss of analytical skills, memory formation, contextual thinking, conversational depth, empathy, and increased anxiety.” TikTok’s executives also knew that compulsive use can interfere with sleep, work and school responsibilities, and even “connecting with loved ones.”
They reportedly knew, as well, that the app’s time-management tool barely helps in keeping young users away from the app. While the tool sets the default limit for app use to 60 minutes a day, teens were still spending 107 minutes on the app even when it’s switched on. That’s only 1.5 minutes shorter than the average use of 108.5 minutes a day before the tool was launched. Based on the internal documents, TikTok based the success of the tool on how it “improv[ed] public trust in the TikTok platform via media coverage.” The company knew the tool wasn’t going to be effective, with one document saying that “[m]inors do not have executive function to control their screen time, while young adults do.” Another document reportedly said that “across most engagement metrics, the younger the user, the better the performance.”
In addition, TikTok reportedly knows that “filter bubbles” exist and understands how they could potentially be dangerous. Employees conducted internal studies, according to the documents, wherein they found themselves sucked into negative filter bubbles shortly after following certain accounts, such as those focusing on painful (“painhub”) and sad (“sadnotes”) content. They’re also aware of content and accounts promoting “thinspiration,” which is associated with disordered eating. Due to the way TikTok’s algorithm works, its researchers found that users are placed into filter bubbles after 30 minutes of use in one sitting.
Advertisement
TikTok is struggling with moderation, as well, according to the documents. An internal investigation found that underage girls on the app were getting “gifts” and “coins” in exchange for live stripping. And higher-ups in the company reportedly instructed their moderators not to remove users reported to be under 13 years old unless their accounts state that they indeed are under 13. NPR says TikTok also acknowledged that a substantial number of content violating its rules get through its moderation techniques, including videos that normalize pedophilia, glorify minor sexual assault and physical abuse.
TikTok spokesman Alex Haurek defended the company and told the organization that the Kentucky AG’s complaint “cherry-picks misleading quotes and takes outdated documents out of context to misrepresent our commitment to community safety.” He also said that TikTok has “robust safeguards, which include proactively removing suspected underage users” and that it has “voluntarily launched safety features such as default screentime limits, family pairing, and privacy by default for minors under 16.”
You must be logged in to post a comment Login