Apple has set a date for its annual September iPhone event, so it’s almost time to see the iPhone 15. All models (save for perhaps a new SE) could ditch the notch and start using the Dynamic Island cutout that debuted in the iPhone 14 Pro and Pro Max. There’s no word on the return of the iPhone mini, and we may have to wait until next year to see the rumored iPhone Ultra.

Apple

Apple may finally phase out the Lightning port with the iPhone 15, making a long-awaited switch to USB-C. The company doesn't have much of a choice there, since all tablets and smartphones sold in the European Union will need to use that charging port starting next year. Apple has also inched towards USB-C on its iPad device family. It might finally be time.

– Mat Smith

​​You can get these reports delivered daily direct to your inbox. Subscribe right here!​​

The biggest stories you might have missed

Meta took down thousands of fake accounts linked to massive Chinese propaganda campaign

German Bionic’s latest exoskeleton helps healthcare workers lift elderly patients

Google’s new routines make it a lot easier for people to automate their smart home

TCL's new budget phones are the first to feature NXTPAPER displays

‘Super Mario Bros. Wonder’ will have its own Nintendo Direct on August 31st

Sony's two new A7C series cameras offer premium features for less money

Foxconn's $10 billion investment promise ended in empty buildings

It's now selling two mostly empty buildings in Wisconsin.

When Foxconn announced its plans to open facilities in Wisconsin back in 2017, it promised to invest $10 billion into bringing production to the US, which was expected to lead to as many as 13,000 jobs. That never came to fruition.

In 2021, Foxconn massively altered the scale of the project and told the local government it would invest $672 million instead of the intended $10 billion. It also reduced the number of potential jobs to 1,454 from 13,000 positions. The Taiwanese supplier is selling two properties in Eau Claire and Green Bay, purchased for almost $12 million in 2018.

Continue reading.

The 'Gran Turismo' movie can't help but be cringe

It’s a decent sports movie stifled by Sony’s marketing machine.

Based on an improbable true story, the movie follows Jann Mardenborough (Archie Madekwe), a 20-something Gran Turismo fanatic who wins a Nissan-sponsored contest to race professionally. Even more improbable, he manages to hold his own in the racing world. The film constantly reminds you it's meant to sell you Sony products in a reality where Apple doesn't exist. At one point, a character is inexplicably attached to his Walkman cassette player, and he only moves on when he's gifted a modern Walkman digital music player. Yeah, you’ll wince.

Continue reading.

Samsung's 85-inch outdoor Terrace TV costs $20,000

Its Neo QLED screen should be ideal for direct sunlight.

Samsung announced a bigger 85-inch Terrace Full Sun set for an eye-watering $20,000. For the money, you'll get a Neo QLED screen, which should remain bright in direct sunlight, as well as Direct-Sun Protection to keep the TV safe. Samsung claims it's protected "up to six hours in sunlight at 700 watts and 104 degrees Fahrenheit," but it may decrease brightness to deal with higher temperatures and sun conditions. The 65-inch Full Sun Terrace is now on sale for a relatively reasonable $6,500, while the 75-inch goes for $9,000.

Continue reading.

Meta's avatars finally grow some legs

They're only available to beta testers for now.

Meta

It's been nearly a year since Meta announced it’d give its metaverse avatars some legs to make them appear slightly more human. Now, Quest Home avatars sport extra limbs in the latest beta version of the Quest software, but you won't see legs on your avatar when you look down, as UploadVR points out. They'll only be visible in third-person or when you're looking at a virtual mirror (much like in many first-person shooter games).

Continue reading.

Garmin is launching the Venu 3 and smaller Venu 3S smartwatches today with a greater focus on sleep-tracking and personalization. In particular, it's adding a proactive sleep coaching feature and nap detection. The latter is noteworthy, considering most modern smartwatches from companies like Apple, Samsung, and Google can track your sleep, but only at night. With the Venu 3 series, you'll be able to rest easy knowing your daytime slumber can also be tracked, if that's your kind of thing.  

The company's new sleep coach considers factors such as a user's daily activity levels, sleep history, and heart rate variability (HRV) – which is a sleep metric new to the Venu series. The coach will provide recommendations on the amount of sleep you’ll need to make it through the next day. The naps that the Venu 3 detects will also go towards replenishing energy levels as tracked by the company’s “Body Battery” tool, which quantifies energy levels on a scale of 1-100 depending on activity expenditure.

Aside from sleep-tracking upgrades, the Venu 3 has a host of other improvements over its predecessor. Notably, it introduces a wheelchair mode that will track pushes rather than steps, as well as handcycle activities and pre-loaded workouts for wheelchair users, offering more inclusivity. The new watch also offers expanded insights on how workouts can affect a user and how much time is needed to recover after every gym session. 

Additionally, Garmin is bringing a “rate of perceived exertion” feature, which enables users to report how challenging a workout felt. With this, the company is trying something new in the fitness arena. Instead of just crunching numbers, it’s dipping its toes into subjective data – the kind of information that's all about personal judgment and individual experiences. The Venu 3 also introduces “interval creation,” which gives a runner or biker the option to build interval workouts from the watch face.

Garmin

Beyond expanding health and fitness tracking features, Garmin says this watch will have a built-in speaker and microphone that will allow a wearer to make calls from the watch. The Venu 3 can also display photo messages on its AMOLED touchscreen. You'll be able to make contactless payments using Garmin Pay and stream music from Spotify and Amazon Music, as well.

The Venu 3 will start at $449.99 and offers up to 14 days of battery life in smartwatch mode. It features a light stainless steel bezel and has silicone bands in either white or black, while five strap color options are available for the smaller 3S.

It’s still unclear if older Venu models will receive software updates that will bring these new features over. But for a mainstream smartwatch from a company that mostly caters to a fitness-focused audience, the Venu 3 appears to be a step forward.

As it promised last week, Samsung has launched Food, a "personalized, AI-powered food and recipe" app in eight languages and 104 countries around the world. It draws on the food database of Whisk, an app Samsung acquired a few years back — and resembles a version of Whisk the company revealed last year. Given Samsung's large presence in kitchens with its smart fridges and other appliances, the release of a food and recipe app seems a logical step for the company.

The app allows users to search for recipes around the world, save them and make weekly eating plans. The company prepared over 160,000 recipes for launch, with that number set to increase down the road. Samsung Food can also be run on mobile phones and Samsung Family Hub smart appliances like refrigerators, while allowing users to manage ingredients, shopping, etc.

Samsung

Users can save recipes anytime, and the app analyzes them, standardizes the format and organizes them to create shopping lists based on the ingredients. It can also provide recipe recommendations based on available food items as managed by the user. It even has a "personalize recipe" function that uses the AI to alter recipes recipes and create vegan or vegetarian versions, for instance. "Users will even be able to create fusion recipes, such as Korean versions of Italian dishes, and adjust cook time or skill level of recipes," Samsung adds.

It uses AI to create recommendations for individualized daily meal plans based on dietary preferences and favorite cuisine types. Nutritional ingredient breakdowns can be viewed at any time, and users can add items to shopping lists and then send them straight to a retailer's e-commerce checkout. With connected cooking, it lets users preheat ovens, set timers and transfer cook settings to supported appliances via a step-by-step guided cooking mode. 

Last week, Samsung said it hoped to secure a million users for the app around the world. While there are numerous recipe apps out there (Mealtime, Paprika, Yummly etc.), Samsung may have an edge with the millions of its smart appliances sold — making it a known quantity to consumers. 

Samsung plans to add new features, like integration with Samsung Health to sync parameters like BMI and calorie consumption, while offering suggestions for diet management. The app will incorporate AI vision tech by 2024, allowing Samsung Food to recognize food items through the camera and provide details like nutrition information. Samsung Food is now available for download on Android and iOS. 

Google has published a photo of the Pixel 8 Pro before it has even unveiled the device... on the day Apple announced that it was going to launch the iPhone 15 on September 12th. An account on X (formerly Twitter) called Android Setting (via The Verge) has spotted a photo of a person holding a beige-colored phone against their ear on a page in the Google Store. We could no longer see the image on the page, which promotes Google's subscriptions and services, but its alt text reportedly read: "A person takes a call on a Pixel 8 Pro phone in Porcelain." 

The device in the image resembles the phone in previously leaked photos of the model's prototype. Its rear cameras are inside one glass panel within a raised metal bar spanning the whole width of the phone. And, of course, Google's "G" logo is at the center of the device. Based on other previous leaks — yes, there were many — the Pixel 8 Pro's camera array is made up of a 50-megapixel main camera that lets in 50 percent more light and a 64-megapixel ultrawide Sony camera. It's also rumored to have a flat display instead of curved ones similar to the Pixel 6 Pro's and Pixel 7 Pro's. In addition, it's expected to come with a 5,000 mAh battery and a 27W max charging speed. 

Google's Pixel 8 Pro doesn't have a launch date yet, but the tech giant unveiled the two previous generations during an event held in October 2021 and 2022. 

It’s been years and years since the bleeding edge of the smartphone world was truly exciting. The industry may want us to salivate over its latest 5G, 108-megapixel, folding-screen masterpiece, but my mouth remains bone dry. To me, it’s because the ceiling of what’s possible hasn’t risen much in the last four or five years, but the floor has maintained its slow creep northward. On paper, the just-announced Fairphone 5 has the specs you’d expect from a mid-range handset. But now that’s not really a penalty since you’re also getting a modular, sustainably-made and potentially upgradeable handset built by people paid a fair(er) wage for doing so. It’s just a smartphone, like any other. On one hand, that makes the new Fairphone 5 boring, but from this perspective, boring can be good.

Ten years into its mission to build a “better” smartphone, it’s not clear if Fairphone still needs a full introduction. For the unaware, the Dutch social enterprise makes devices that are “fairer” than the competition. That means paying factory workers a living wage, sourcing “ethical” raw materials, using recycled materials and keeping the hardware out of the trash for as long as possible. To that end, each device is modular, easy-to-repair, sometimes upgradeable and often far longer lasting than its competition. It’s less fancy than, say, Google’s long-mourned Project Ara, but it’s a damn sight more real than that imagined dream of a modular, upgradeable phone. Here, it’s not just possible, but easy, to pick up a Philips 00-head screwdriver and fit a new component in a matter of minutes.

The Fairphone 5 doesn’t stray far from the template laid down by its immediate predecessor in size and style. It has the same hefty chassis intercut with antenna lines, the same camera housing, the same fingerprint-sensing power button and the same easily removable backplate. Like a lot of phones released in 2023, the changes are nips and tucks to reflect where the industry is at. So, the battery is more capacious, the display a little bigger, and the cameras are a hair more powerful now than they were a generation ago. Fairphone has also made it possible to swap out each camera separately, making it easier and cheaper to fix or upgrade one down the line.

You’ll be able to grab plenty of spare parts, with only the main chassis frame not being available separately. The product list includes the 90Hz, 6.46-inch LG-made POLED display, 4,200mAh battery and all three cameras. You can also get a spare earpiece, loudspeaker, USB-C port and a top unit containing the Time of Flight sensor, as well as the SIM and SD-card slots. If a part on this phone breaks, you’ll be able to pick up almost everything you could need for not that much cash. The priciest component is the display, priced at €99.95 (around $108), which is about half what Samsung charges, and almost a third of what Apple asks for.

To make good on its promises of sustainability and longevity, Fairphone 5 uses Qualcomm’s octa-core QCM6490 SoC. It’s an industrial-grade IoT chipset that, far as I can tell, is used in only one other phone: AGM’s thermal-camera equipped G2 Guardian. There’s only one SKU here, with the Fairphone 5 coming with 8GB RAM and a very generous 256GB storage, expandable to 2TB with the right microSD card. Fairphone says the choice was dictated by the chipset’s blend of premium features like 6G, WiFi 6E and on-chip AI processing. Not to mention Qualcomm promises to support its industrial grade silicon for longer than it does its mobile chips. To that end, Fairphone is committing to support the (almost) stock install of Android 13 and “at least five operating system upgrades after.” (Given Android 14’s forthcoming launch, you should expect that to be heading to your phone in the near future.) The company says it expects to offer software support until 2028 at the very least, but hopes it’ll keep things running until 2033.

And Fairphone has now established a track record of being able to keep its phones going for a fair old while. The Fairphone 2 was launched in 2015 with Android 5.1 and got its final security patch in March 2023 with an Android 10 update. Seven or so years of software support is a good reason to switch, especially when Samsung pledges four years of software and five years of security updates for “select” handsets.

Photo by Daniel Cooper / Engadget

In my limited time playing with the new model, I’ve found the performance to be a lot snappier than on older Fairphones. Animations and transitions in the OS itself don’t appear at this early stage to be laggy, and Fortnite on Epic mode played perfectly at 30fps. I certainly noticed how much faster it is to unlock this phone compared to the last one, both using my face and my fingerprint. At first blush, this appears to be more than capable of meeting your expectations for normal phone use. Well, if your idea of normal phone use is texting, emails, social media, YouTube, TikTok and the odd deathmatch.

I’d describe the Fairphone 5’s imaging as solid, although I’m worried that to do so is to damn it with faint praise. The primary 50-megapixel, f/1.88, OIS-equipped camera backed by Sony’s IMX800 sensor is more than enough for most people’s photography needs. It may not have the software smarts of some devices in its class, but it’s no slouch, especially with video. Glance at the Fairphone 5’s back and you’ll spot a circle the same size as the camera lenses, which is the ToF sensor. It’s this phone’s secret weapon, especially when it comes to the lightning-fast autofocus when you’re shooting video.

It’s not all good: The phone gets a little warmer than I’d like when running heavy load, but that feels par for the course in the 5G era. The built-in speakers, meanwhile, are so tinny I’d never recommend using this for audio if you have headphones to hand. I also think the solid black cover will be far better at resisting smears than the transparent edition in my loan unit. And I wish, as well as the black, transparent and sky blue colorways, the company had offered the same green and orange paint job that is so wonderful on the Fairbuds XL.

It’s also a shame we’re not yet seeing a wide release for the Fairphone 5 in the US, at least not at the same time as it comes to Europe. The Fairphone 4 is available to buy in the States with a “deGoogled” OS courtesy of privacy-focused startup Murena. I’m sure the 5 will make a similar journey in future, but it’d be nice to see a simultaneous release so people aren’t left waiting for a year or more to get the updated hardware.

I’ve said before that Fairphone has always carried the whiff of a compromise choice, the cork and hemp sneaker you buy to assuage your guilt over all the sweatshop-made kicks you own. But, both thanks to Fairphone’s efforts to improve its product, and the general stagnation in the mobile industry more generally, the delta between what’s hot and what’s not has closed by a lot. If you’re looking to pick up a mid-range handset that you can keep going for twice as long as any other phone on the market, and you want to do a bit of good in the process, this is probably the phone for you. It’s not often the view from the moral high ground is this comfortable.

The Fairphone 5 is available to pre-order today, August 30, from the company’s website and a number of major European carriers. It is priced at €699 (around $750), with shipping expected to commence September 14 from Fairphone and the majority of its carriers.

Sony has just split its small form-factor full-frame A7C series into two, effectively offering lower-cost versions of higher-end cameras — much as it did with the ZV-E1 vlogging camera. The new models are the 33-megapixel A7C II (based on the A7 IV), and the 61-megapixel A7C R, a mini-me version of the A7R V. Both are nearly as powerful as the larger versions and offer key ergonomic updates, but lack features that many pros find essential. 

Sony A7C R

We'll start with the A7C R, since it's an all-new model. It features a softer, grippier material along with a deeper handle. It also comes with an add-on grip that extends the length, giving you a place to put your pinky finger — a nice addition that unfortunately also makes the camera less compact. 

Sony

Still, those are key improvements that address ergonomic issues I've noticed on past Sony models — particularly when using them for a long day of shooting. Sony also added a front dial so you can more easily adjust aperture/shutter speed/exposure compensation, along with a dedicated video/photo/S&Q dial. And of course, the A7C R weighs just 515 grams (18.2 ounces) compared to the A7R V's 723 grams (25.5 ounces) and is nearly half the size. That makes it a better choice for street photography and more. 

It has a flip-out one-million dot display and 2.36-million dot OLED electronic viewfinder with slightly improved .70X magnification over the A7C II. It also offers full in-body stabilization with 7.0 stops of compensation (depending on the lens) — a nice feature in such a small camera. Other features include mic/headphone ports, USB-C 3.2, a micro HDMI connector a and single UHS-II card slot. The latter feature is a show-stopper for many pros, as there's no backup for potentially priceless photos. 

Sony

In terms of photo specs, it's close to the A7R V with a few notable differences. You can shoot 61-megapixel RAW images at up to 8fps, compared to 10fps on the A7R V. The A7C R has a smaller buffer, though, and the single lower-speed SD UHS-II card slot (the A7R V has dual high-speed CFexpress Type A and SD UHS II slots), so you won't be able to capture nearly as many photos in a burst. Like the A7R V, the A7C R has 693 autofocus points and 79 percent coverage. It uses the same processor, so should offer roughly the same AF performance in terms of burst shooting, AI tracking and more.

Most importantly, you can expect identical image quality to the A7R V. As I said in my review of the latter, with the very high resolution and the lack of an anti-aliasing filter, only Hasselblad and Fuji’s 100-megapixel medium format cameras offer greater detail. It also offers Sony’s Pixel Shift Multi-Shot that can quadruple resolution to 240.8 megapixels.

Sony

The A7C R can't shoot 8K video like the A7R V, likely due to heating issues with the smaller body. However, it still offers 4K 60p full-frame video (with pixel binning) and 6.2K oversampled 1.5x cropped 4K 60p video. It comes with S-Log3, S-Cinetone and S-gamut3 Cine, records 10-bit 4:2:2 sampled video and can shoot at up to 120 fps in 1080p mode. Data rates top out at 600Mbps in XAVC S-I 4K mode, but you'll need a (pricey) V90 SD card to shoot that. 

That takes us to a very important consideration: the price. The A7C R costs $3,000, or a full $900 less than the A7R V — something that will give many buyers extreme pause. It arrives in fall of 2023. 

Sony A7C II

As mentioned, the A7C II has the same 33-megapixel sensor and many of the same specs as the A7 IV. In terms of the body and handling, the form factor and weight are identical to the A7C R — so you get the same extra dial, updated grip, improved EVF and other previously-mentioned features. Unlike the A7C R, though, it doesn't come with the add-on grip extension in the box.

Shooting speeds are a bit faster at 10 fps in mechanical or electronic shutter mode. By the way, neither the A7C II or A7C R have full mechanical shutters, but use electronic first curtain shutters. This can create bokeh issues when using lenses with very fast apertures, but otherwise isn't noticeable. 

Sony

What's interesting about the A7C II is that autofocus is superior to the A7 IV because it has the same dedicated AI processing unit as the A7C R (and A7R V) that's missing on the A7 IV. That means you can expect improved subject detection, tracking and other features compared to the A7 IV. When compared to the A7C, though, Sony promises a 40 percent improvement for animal and bird AF tracking, 60 percent for eye AF and 20 percent for auto exposure tracking. 

Again, image quality should be the same as the A7 IV. That latter delivers sharper photos than the A7 III thanks to the extra resolution, but offers superior low-light capability despite the smaller pixels. A downside to the A7 IV is rolling shutter, so expect the same issue on the A7C II. 

Sony

In terms of video, the A7C II is actually a better option than the A7C R. It offers sharp full-frame 4K 30p video with 7K oversampling, or 4K 60p with a 1.5x crop. You get 10-bit 4:2:2 color sampling with S-Log3 and S-Cinetone, with LUTs available in camera. 

The Sony A7C II is available in silver and black, and arrives this fall for $2,200. That price is a significant $300 less than the A7 IV and it's actually a better camera in some ways — provided the lack of a second card slot doesn't bother you. 

A pair of popular OnePlus smartphones just went on sale, hitting record low prices for both. The company’s flagship OnePlus 11 5G went down from $700 to $600, a savings of nearly 20 percent. The budget-friendly OnePlus Nord N30 5G got even, well, friendlier with a $50 discount, dropping the cost to $250 from $300. If you’re shopping for a smartphone, this is a good time to take the plunge.

We praised the OnePlus 11 as a “back-to-basics flagship smartphone,” noting its gorgeous 120Hz 6.6-inch OLED display, the fantastic battery life, 100W quick-charging and improved camera system when compared to its predecessor. In other words, the 11 was already a bargain at $800, as modern iPhones and Samsung phones cost upwards of $1,000. Today’s sale makes the bargain even harder to resist.

The OnePlus Nord N30 takes a more modest approach, as this is absolutely a low-priced smartphone rather than a flagship. However, it’s one of the best budget-friendly phones around and a great choice for anyone looking for a no-frills device that gets the job done. The specs are fantastic for the price, with a Snapdragon 695 processor, 8GB of RAM, 128GB of storage and a crisp 120Hz IPS display. Not many cheap phones can match this set of features.

These phones aren’t perfect, as the N30 lacks waterproofing and the 11 isn’t the most exciting flagship model in the world, but the list of pros far outweigh any list of cons. OnePlus isn’t widely available at retail outlets, so this sale is reserved for Amazon and Best Buy.

Follow @EngadgetDeals on Twitter and subscribe to the Engadget Deals newsletter for the latest tech deals and buying advice.

General Motors is taking Google’s AI chatbot on the road. The automaker announced today that it’s using Google Cloud’s Dialogflow to automate some non-emergency OnStar features like navigation and call routing. Crucially, the automaker claims the bot can pinpoint keywords indicating an emergency situation and “quickly route the call” to trained humans when needed. GM says the system frees up OnStar Advisors to spend more time with customers requiring a live human.

According to GM, the OnStar Interactive Virtual Assistant (IVA) has used Google Cloud’s Dialogflow under the hood since IVA’s 2022 launch. The virtual voice assistant can handle common customer questions and help with routing and navigation, including turn-by-turn directions. The companies see the collaboration as expanding down the road. “The successful deployment of Google Cloud’s AI in GM’s OnStar service has now opened the door to future generative AI deployments being jointly piloted by General Motors and Google Cloud,” the companies wrote in a joint press release.

The automaker says Google Cloud’s AI has allowed OnStar to better understand customer requests on the first try. In addition, it says customers have reacted positively to avoiding hold times as they can quickly begin chatting with an AI-powered bot with a “modern, natural sounding voice.” GM says the virtual assistant now handles over one million customer inquiries per month in the US and Canada. OnStar IVA is available in most GM vehicles, 2015 and newer, with OnStar connections.

GM has also reportedly worked on developing a ChatGPT-powered assistant for its vehicles, although it isn’t yet clear if that project is still on the table.

“Generative AI has the potential to revolutionize the buying, ownership, and interaction experience inside the vehicle and beyond, enabling more opportunities to deliver new features and services,” Mike Abbott, GM’s executive vice president of software and services, wrote in the press release. “Our software-led approach has accelerated the creation of compelling services for our customers while driving increased efficiency across the GM enterprise. The work with Google Cloud is another example of our efforts to transform how customers engage with our products and services.”

The companies also announced today that Google’s Dialogflow tech is behind chatbots on the GM website, similar to the slew of OpenAI-powered assistants that began popping up since the launch of the ChatGPT API earlier this year. GM’s web bots can “conversationally help answer customer questions about GM vehicles and product features based on the technical information from GM’s extensive vehicle data repositories,” according to the automaker.

“General Motors is at the forefront of deploying AI in practical and effective ways that ultimately create better customer experiences,” Thomas Kurian, Google Cloud CEO, wrote today. “We’re looking forward to a deepened relationship and more collaboration with GM as we explore how the company uses generative AI in transformational ways.”

Several months after WhatsApp released a Windows desktop client, Mac users are getting to join the party with their own dedicated app for the service. The formal arrival of the client (which had been in beta since January) on Apple's desktops and laptops means users can take part in WhatsApp group calls on their Mac for the first time. 

WhatsApp for Mac supports up to eight people in video calls and as many as 32 in audio-only chinwags. You can hop into a group call after it's already started, view your call (and chat) history and opt to receive notifications about incoming calls even if you don't have the WhatsApp client open. Sharing files should be a cinch too, as you'll be able to simply drag and drop them into a conversation.

The WhatsApp team has spent quite some time making sure that the service supports end-to-end encryption (E2EE) across multiple devices with cross-platform syncing. So, it's not super surprising that WhatsApp for Mac includes E2EE protection for your chats and calls. The app is available from the WhatsApp website and it'll hit the Mac App Store soon.

Google took a step towards transparency in AI-generated images today. Google DeepMind announced SynthID, a watermarking / identification tool for generative art. The company says the technology embeds a digital watermark, invisible to the human eye, directly onto an image’s pixels. SynthID is rolling out first to “a limited number” of customers using Imagen, Google’s art generator available on its suite of cloud-based AI tools.

One of the many issues with generative art — apart from the ethical implications of training on artists’ work — is the potential for creating deepfakes. For example, the pope’s hot new hip-hop attire (an AI image created with MidJourney) going viral on social media was an early example of what could become more commonplace as generative tools evolve. It doesn’t take much imagination to see how something like political ads using AI-generated art could do much more damage than a funny image circulating on Twitter. “Watermarking audio and visual content to help make it clear that content is AI-generated” was one of the voluntary commitments that seven AI companies agreed to develop after a July meeting at the White House. Google is the first of the companies to launch such a system.

Google doesn’t go too far into the weeds about SynthID’s technical implementation (likely to prevent workarounds), but it says the watermark can’t be easily removed through simple editing techniques. “Finding the right balance between imperceptibility and robustness to image manipulations is difficult,” the company wrote in a DeepMind blog post published today. “We designed SynthID so it doesn’t compromise image quality, and allows the watermark to remain detectable, even after modifications like adding filters, changing colours, and saving with various lossy compression schemes — most commonly used for JPEGs,” DeepMind’s SynthID project leaders Sven Gowal and Pushmeet Kohli wrote.

Google DeepMind

The identification part of SynthID rates the image based on three digital watermark confidence levels: detected, not detected and possibly detected. Since the tool is embedded into the image’s pixels, Google says its system can work alongside metadata-based approaches, like the one Adobe uses with its Photoshop generative features, currently available in an open beta.

SynthID includes a pair of deep learning models: one for watermarking and the other for identifying. Google says the two trained on diverse images, culminating in a combined ML model. “The combined model is optimised on a range of objectives, including correctly identifying watermarked content and improving imperceptibility by visually aligning the watermark to the original content,” Gowal and Kohli wrote.

Google acknowledged that it isn’t a perfect solution, adding that it “isn’t foolproof against extreme image manipulations.” But it describes the watermark as “a promising technical approach for empowering people and organisations to work with AI-generated content responsibly.” The company says the tool could expand to other AI models, including those tasked with generating text (like ChatGPT), video and audio. 

Although watermarks could help with deepfakes, it’s easy to imagine digital watermarking turning into an arms race with hackers, with services that adopt SynthID requiring continual updating. In addition, the open-source nature of Stable Diffusion, one of the leading generative tools, could make industry-wide adoption of SynthID or any similar solution a tall order: It already has countless custom builds that can run on local PCs out in the wild. Regardless, Google hopes to make SynthID available to third parties “in the near future” to at least improve AI transparency industry-wide.