Connect with us

Technology

OpenAI researchers develop new model that speeds up media generation by 50X

Published

on

OpenAI researchers develop new model that speeds up media generation by 50X

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More


A pair of researchers at OpenAI has published a paper describing a new type of model — specifically, a new type of continuous-time consistency model (sCM) — that increases the speed at which multimedia including images, video, and audio can be generated by AI by 50 times compared to traditional diffusion models, generating images in nearly a 10th of a second compared to more than 5 seconds for regular diffusion.

With the introduction of sCM, OpenAI has managed to achieve comparable sample quality with only two sampling steps, offering a solution that accelerates the generative process without compromising on quality.

Described in the pre-peer reviewed paper published on arXiv.org and blog post released today, authored by Cheng Lu and Yang Song, the innovation enables these models to generate high-quality samples in just two steps—significantly faster than previous diffusion-based models that require hundreds of steps.

Advertisement

Song was also a leading author on a 2023 paper from OpenAI researchers including former chief scientist Ilya Sutskever that coined the idea of “consistency models,” as having “points on the same trajectory map to the same initial point.”

While diffusion models have delivered outstanding results in producing realistic images, 3D models, audio, and video, their inefficiency in sampling—often requiring dozens to hundreds of sequential steps—has made them less suitable for real-time applications.

Theoretically, the technology could provide the basis for a near-realtime AI image generation model from OpenAI. As fellow VentureBeat reporter Sean Michael Kerner mused in our internal Slack channels, “can DALL-E 4 be far behind?”

Faster sampling while retaining high quality

In traditional diffusion models, a large number of denoising steps are needed to create a sample, which contributes to their slow speed.

Advertisement

In contrast, sCM converts noise into high-quality samples directly within one or two steps, cutting down on the computational cost and time.

OpenAI’s largest sCM model, which boasts 1.5 billion parameters, can generate a sample in just 0.11 seconds on a single A100 GPU.

This results in a 50x speed-up in wall-clock time compared to diffusion models, making real-time generative AI applications much more feasible.

Reaching diffusion-model quality with far less computational resources

The team behind sCM trained a continuous-time consistency model on ImageNet 512×512, scaling up to 1.5 billion parameters.

Advertisement

Even at this scale, the model maintains a sample quality that rivals the best diffusion models, achieving a Fréchet Inception Distance (FID) score of 1.88 on ImageNet 512×512.

This brings the sample quality within 10% of diffusion models, which require significantly more computational effort to achieve similar results.

Benchmarks reveal strong performance

OpenAI’s new approach has undergone extensive benchmarking against other state-of-the-art generative models.

By measuring both the sample quality using FID scores and the effective sampling compute, the research demonstrates that sCM provides top-tier results with significantly less computational overhead.

Advertisement

While previous fast-sampling methods have struggled with reduced sample quality or complex training setups, sCM manages to overcome these challenges, offering both speed and high fidelity.

The success of sCM is also attributed to its ability to scale proportionally with the teacher diffusion model from which it distills knowledge.

As both the sCM and the teacher diffusion model grow in size, the gap in sample quality narrows further, and increasing the number of sampling steps in sCM reduces the quality difference even more.

Applications and future uses

The fast sampling and scalability of sCM models open new possibilities for real-time generative AI across multiple domains.

Advertisement

From image generation to audio and video synthesis, sCM provides a practical solution for applications that demand rapid, high-quality output.

Additionally, OpenAI’s research hints at the potential for further system optimization that could accelerate performance even more, tailoring these models to the specific needs of various industries.


Source link
Continue Reading
Advertisement
Click to comment

You must be logged in to post a comment Login

Leave a Reply

Technology

AI networking startup Boardy raises $3M pre-seed

Published

on

AI networking startup Boardy raises $3M pre-seed

Boardy, a professional networking startup driven by AI voice technology, announced Thursday the closing of a $3 million pre-seed round. 

The company was co-founded by its CEO Andrew D’Souza, and brothers Ankur Boyed and Abhinav Boyed. They came up with this idea in March, started building it throughout the summer, and just launched officially this month. 

The way it works is simple: a user gives their number to Boardy.ai and receives a phone call from an AI voice assistant named, of course, Boardy. The person chats to Boardy, telling the AI what they are working on. Boardy then checks if anyone in the Boardy network might be able to help. The network Boardy knows — which D’Souza says consists right now of a few thousand — started with D’Souza’s own network of investors, founders, and creators, and has expanded since then. It is mainly used for people who are looking to meet customers and investors, and has also helped people get into accelerator programs as well as with co-founder matching, he said. 

“If Boardy has spoken with someone he thinks would make a good connection based on both experience, as well as whether the two of you would actually get along, he will try and facilitate a double-opt-in introduction,” D’Souza explained. If the introduction is accepted, then Boardy introduces both parties via email. “You can call Boardy back every week to work on a new introduction for you.” 

Advertisement

D’Souza said they started the company because of how lonely social media has made people. In fact, studies are now showing that America in particular is in the midst of a loneliness epidemic, which started even before the pandemic. D’Souza said there is a fear that AI will exacerbate the loneliness epidemic, taking jobs and displacing what makes people feel human. While other startups are building AI-generated companions, sometimes with disturbing results, Boardy is using AI to facilitate human connections.

“We built Boardy to create a better future, where AI actually makes us more connected to each other and where humans and AI collaborate to solve humanity’s hardest problems,” D’Souza said

Before this, D’Souza co-founded and led the e-commerce company Clearco. After almost ten years at Clearco, he said the company grew to a size where they needed a more seasoned capital markets expert to lead the company. He willingly decided to leave as they brought on a new CEO, while D’Souza set forth on a new path. 

Fundraising for Boardy was easy as the round primarily consisted of investors D’Souza met through Clearco. HF0 was the largest investor in the round, with others including 8VC, Precursor, Afore, FJ Labs, and NextView.

Advertisement

“Going forward, I hope to meet more of my investors through Boardy,” he said.

Boardy will use the fresh capital to continue building and training the AI, hoping to make it smarter and more empathetic. The team is also working to expand Boardy’s personal network to connect users with more people.

There aren’t many competitors to Boardy at the moment, though there are companies building in the AI social networking space, such as Butterflies and SocialAI. There are AI companies to help consumers build agents and help with consumer interactions and booking appointments, though. D’Souza hopes Boardy is different, saying the AI agent “works for himself.” 

“You can ask Baordy for help and he’ll do his best to help you, but not at the expense of other people in his network,” he continued. “You can’t tell Boardy what to do, which is actually what makes him more trustworthy.” 

Advertisement

Source link

Continue Reading

Technology

Tesla has been testing a robotaxi service in the Bay Area for most of the year

Published

on

Tesla has been testing a robotaxi service in the Bay Area for most of the year

Tesla has been testing a robotaxi service in the Bay Area for the past few months, Elon Musk said during the company’s earnings call Wednesday.

The company’s employees have been able to summon an autonomously operated Tesla vehicle for trips using the company’s prototype ride-hailing app, Musk said. The vehicles arrive with safety drivers behind the wheel, ready to intervene in case anything goes wrong.

But Musk said the vehicles are operating autonomously using the latest version of the company’s Full Self-Driving software, which he said will be “1000 times better” than human driving by the second quarter of 2025. And he said he expects to roll out a paid ridehailing service in California and Texas starting next year, pending regulatory approval.

Tesla is not currently licensed to operate a commercial autonomous ridehailing service in California. Musk predicted it would be easier to obtain permission in Texas than California.

Advertisement

To be sure, the current iteration of FSD is a Level 2 driver assist system, which is not autonomous and requires constant human supervision. Musk has promised that FSD will become “unsupervised” next year, but his past predictions about autonomy have generally failed to come to fruition.

Still, the fact that Tesla has been testing its ridehailing function with employees proves the company is still dead set on eventually launching the long-promised Tesla Network. First announced in Musk’s Master Plan Part Deux, the Tesla Network claims to allow regular Tesla owners to send their vehicles out autonomously to function as robotaxis while their owners stay at home.

“This really is a profound change,” Musk said. “Tesla will become more than a vehicle and battery manufacturer company at that point.”

During the earnings call, Tesla executives described certain functions in the current Tesla app, like profile sharing and synchronizing setting across different vehicles, as laying the groundwork for an eventual robotaxi service.

Advertisement

The company first teased several screenshots of a ride-hailing function in its app earlier this year. The first screen shows a big button that says “Summon” with a lower message for the possible wait time. The next screen has a 3D map with a little virtual vehicle following a route to the waiting passenger. It looks a lot like the Uber app — but more Tesla-y.

Source link

Continue Reading

Technology

Got a spare $23,000? Devialet just launched a super slimline streaming amp, and it’s gorgeous

Published

on

Devialet Astra with gold-leaf being applies, on beige background

Want a svelte streaming amp that looks for all the world like it was hewn by the Greek God Chrysus – the one most associated with precious metal? Devialet can help, with its all-new Astra (which is essentially the name of the Titan God of the stars, planets and so on).

The Astra Opéra de Paris edition is finished in 23-carat gold leaf by artisans from Ateliers Gohard. But under its stunning exterior, it is of course built on past greatness. In this case, Devialet wouldn’t necessarily point you towards its spherical Devialet Mania (also available in a ‘sunset rose’ hue) or the company’s fabulous high-end Dione soundbar which boasts the same drivers as those premium in-flight seats. No, Astra is the evolution of Devialet’s first creation, the Devialet Expert amplifier, unveiled 15 years ago – yes, now I also feel as old as the Gods.

Source link

Continue Reading

Technology

Garmin launches new Fenix 8 series smartwatches in India; price starts at Rs 86,990- The Week

Published

on

Garmin launches new Fenix 8 series smartwatches in India; price starts at Rs 86,990- The Week

Garmin smartwatches, known for its high-quality GPS, fitness stats, durability, and detailed workout features, has released its new Fenix 8 series in India. The smartwatch has already garnered an overwhelming response from the global market.

The Fenix 8 series is available in three sizes – 43mm, 47mm, and 51mm. While all come in the AMOLED version, the solar-charging models are available only in the 47mm and 51mm variants. The watch is topped with a scratch-resistant sapphire lens and features a titanium bezel. The additional features of the multi-sport smartwatch include a built-in speaker and mic for voice commands and phone calls. However, the highlight of the new launches is the fitness and workout features that allow for exhaustive performance tracking.

ALSO READ | Redmi Watch 5 Active: Chunky budget-friendly smartwatch with long battery life and bright display

There are advanced strength training plans for sport-specific workout activity trackers; advanced mapping with TopoActive maps with relief shading and built-in maps for golf courses and ski resorts worldwide; dynamic round trip routing to help wearers set how far they want to go and for guiding them back on time; ski difficulty tracking to let the user know how much time they are spending on different difficulties throughout the day; and dive capability with a 40-metre dive rating and leak-proof buttons. The ‘Garmin share’ feature allows the wearer to easily share saved locations, courses, and workouts with friends, while the Garmin messenger app helps one communicate directly via their smartwatch. Additionally, the brand has also announced a personalised application, designed specifically for coaches and athletes to map their performance.

Advertisement

“The Fenix 8 Series is perfect for individuals looking to elevate their performance or maintain a healthy lifestyle. Whether you are an experienced athlete or someone who enjoys staying active, this new series offers advanced features and extended battery life for outstanding versatility,” says, Tim Spurling, General Manager, Emerging Markets CAMEA, Garmin.

Along with these, the watch is packed with a built-in LED flashlight, and 24/7 health and wellness trackers, which include wrist-based heart rate, advanced sleep monitoring, respiration tracking, pulse, and more. There is music control, a connect IQ store for adding and changing watch faces and data field, endurance score, body’s energy level tracker, and much more. The AMOLED display model runs up to 29 days, while the solar charging variants have a life of up to 48 days. The watches start at Rs. 86,990.

Source link

Continue Reading

Technology

Qualcomm made the future of smartphone cameras a lot more exciting

Published

on

Qualcomm made the future of smartphone cameras a lot more exciting

Qualcomm made big announcement this week. The company just unveiled its new Snapdragon 8 Elite chip, and even if you don’t keep a particularly close eye on the smartphone chipset world, it’s something that’s worth getting excited about. Qualcomm is promising substantial performance and efficiency improvements over last year’s already excellent Snapdragon 8 Gen 3, which is great news for next year’s slate of flagship Android phones.

But there’s more to the Snapdragon 8 Elite than it being more powerful and more efficient. It also has the potential to substantially change the way we use the cameras on our phones. How so? I talked to Judd Heape, VP of product management at Qualcomm, to better understand it myself, and I came away significantly more excited about the immediate (and faraway) future of our smartphone cameras.

Behind-the-scenes camera upgrades that matter

Qualcomm Snapdragon 8 Elite powering a phone.
Qualcomm

A big part of the Snapdragon 8 Elite is what Qualcomm calls its “AI ISP.” An ISP (image signal processor) is a standard component of every smartphone chip, including Snapdragon ones, and is what facilities image processing on your phone’s camera. For the Snapdragon 8 Elite, the AI ISP has a much tighter connection to the NPU (neural processing unit) than any other Snapdragon chip before it. That may sound like a lot of technical jargon, but it essentially means that critical camera features should run much better than before.

What kind of camera features? One of the most exciting is how the AI ISP should improve auto white balance. Why is that a big deal? “One of the things that cameras get wrong a lot based upon very complex lighting, like lighting an interior versus exterior at the same time, if you’re outdoors in a parking lot and the lights are very orange … that sort of thing,” Heape said. “Your skin tone can get really messed up really easily because of that, and that’s a failure of auto white balance.”

Because this new ISP has a tighter connection to the Snapdragon 8 Elite’s NPU, it can “generate proper skin tone no matter what the lighting condition is.” The really exciting thing is that these auto white balance improvements don’t just happen after you’ve taken a photo. You see those enhancements in real time through the viewfinder, so the image you see as you’re taking a picture is what you’ll get.

Advertisement

Another promising change is that the new ISP consumes less power. “The power consumption also of the ISP has gone down due to the new architecture,” Heape said. “So, in really interesting use cases, like 4K 60 fps HDR video recording, the ISP consumes 25% less power … which means that you’ll have less thermal problems when you’re trying to shoot videos.” Speaking of video, the new ISP also improves Qualcomm’s “temporal noise filters.” Those filters look at more video frames than before so “the video that you’ll shoot is a lot cleaner than it used to be. It was good before but it’s even better now.”

Are these flashy AI camera features like the Google Pixel 9’s Add Me mode or the numerous camera/photo editing tools in Galaxy AI? No. But are they ones that could legitimately result in better photos and videos for any phone with a Snapdragon 8 Elite? Absolutely. And that’s the type of AI camera enhancements I want to see more of.

What does the future of smartphone cameras look like?

A person taking the Xiaomi 14 Ultra out of a pocket.
Andy Boxall / Digital Trends

Beyond immediate enhancements coming with the Snapdragon 8 Elite, Heape also shared a few insights about what the greater future of smartphone cameras may look like — and what he, as someone who works closely on this stuff at Qualcomm — wants to see more of.

While talking about the auto white balance improvements and seeing those enhancements through the viewfinder in real time, Heape admitted that “the industry needs to drive toward that. What you see is what you get is really important. It gives the photographer confidence … OEMs need to concentrate on that.”

Heape was also asked about his “dream application” for Qualcomm’s ISP advancements and what he was most excited to see smartphone companies do with it in the next few years — and I thought his answer was fascinating. As he explained, Heape is interested in “reducing the cost and complexity of the camera system.”

Advertisement

“I think we can get away with two cameras instead of three in some cases … which reduces the processing and power … plus using AI for super resolution and using AI to augment capabilities in lowlight.” When asked to look even further ahead about how he’d like to see AI used to further improve the smartphone camera experience compared to what we have today, Heape said he wants to see a world where your camera gets to know you.

Close-up picture of the three rear cameras on the Samsung Galaxy S24 FE.
Joe Maring / Digital Trends

“Your camera getting to know you, and know what you like, and know the edits you tend to make and the shots you like to take … basically training your camera. Learning what kind of shots you like and the composition and the camera getting to know you over time and then making those adjustments for you the more images you shoot. I think that’s kind of where we need to get to next … kind of like having the Copilot PC, if you will, for your camera.”

As someone who’s felt pretty unimpressed with existing AI camera tools, I really hope Heape’s ideal camera future is the one we’re headed toward — one where AI is working in the background and giving you better photos and videos without you having to think about it. I don’t particularly care about (or want) AI features that alter my photos into something they aren’t. I want my phone to take the best picture possible without me needing to think about it too much, and talking with Heape, that sounds like the future he wants to see, too.

I think we’re heading in the right direction

A person taking a photo with the OnePlus 12.
Andy Boxall / Digital Trends

Ever since smartphone brands and chip manufacturers started going all in on AI over the last couple of years, I’ve found it difficult to get truly excited about almost any of it. We’ve seen cool tech demos and a few cool features here and there, but nothing that I’ve felt has genuinely changed how we use our phones — particularly when it comes to the camera.

While it remains to be seen just how well the Snapdragon 8 Elite and its new ISP actually perform in the real world — and whether the camera future Heape describes is the one we’re actually headed toward — I will admit that I’m genuinely curious and hopeful about all of it. I firmly believe that the best use of AI is having it work in the background and allowing you to use your phone as you normally would but making it better. Give me better white balance and video recording any day of the week over wonky image generator tools. It really feels like that’s the direction Qualcomm is headed, and if that’s the future we can look forward to with smartphone cameras, count me in.






Source link

Advertisement

Continue Reading

Technology

Xiaomi 15 Ultra appears with odd-looking camera layout

Published

on

Featured image for Xiaomi 15 Ultra appears with odd-looking camera layout

The Xiaomi 15 Ultra has just surfaced, and it has an odd-looking camera layout. These images comes from SmartPrix, a site that partnered with Yogesh Brar, a tipster.

Do note that the tipster in question here has a mixed track record, but this is not the first time this design has appeared. It surfaced earlier this month in a sketch form, and this is only its render form.

The Xiaomi 15 Ultra has just appeared with an odd-looking camera layout, but these are not official renders

It’s obvious that the renders included here have been created by a third-party, they’re not official leaks. So it’s possible they’re simply based on the sketch that surfaced earlier this month.

In any case, you will notice that the phone has rounded corners and a centered display camera hole. The bezels are very thin, and even though they don’t look uniform in this render, chances are that they will be.

Advertisement

All the physical buttons sit on the right-hand side of the phone. We don’t see its frame on the side, but it’ll likely be mostly flat. On the back, you’ll spot a huge camera oreo, which takes up a huge chunk of the phone’s upper side.

Inside that camera oreo, four cameras are located. The main one sits at the bottom, and it’s centered between two other shooters. The periscope telephoto camera is included in the top-right corner.

Leica will once again be a part of the package

The Leica logo is also visible on the inside of this camera oreo, and the same goes for an LED flash. The camera island shown here will likely protrude quite a bit on the back of the device. Xiaomi will use larger camera sensors.

The source pictured this phone in black and white colors here, with a vegan leather backplate., It’s quite possible that the phone will arrive in those colors, with that material on the back. Just note that these are not leaked renders, so these are not color leaks either. The real thing will likely look at least a bit different.

Advertisement

Unlike the Xiaomi 15 and Xiaomi 15 Pro, the Xiaomi 15 Ultra is not expected to arrive this month. Those two phones are coming on October 29, while the Xiaomi 15 Ultra will follow next year.

Source link

Continue Reading

Trending

Copyright © 2024 WordupNews.com