Adobe’s Project Sweet Talk makes portraits come alive

One of the most interesting sessions at Adobe MAX is traditionally the Sneaks keynote, where engineers from the company’s various units show off their most cutting-edge work. Sometimes, those turn into products. Sometimes they don’t. These days, a lot of the work focuses on AI, often based on the Adobe Sensei platform. This year, the company gave us an early look at Project Sweet Talk, one of the featured sneaks of tonight’s event.

The idea here is pretty straightforward, but hard to pull off: take a portrait, either a drawing or a painting, identify the different parts of the face, then animate the mouth in sync with a voice-over. Today, Adobe’s Character Animator (which you may have seen on shows like The Late Show with Stephen Colbert) does some of that, but it’s limited in the number of animations, and the result, even in the hands of the best animators, doesn’t always look all that realistic (as far as that’s possible for the kind of drawings you animate in the product). Project Sweet Talk is far smarter. It analyzes the voice-over and then uses its AI smarts to realistically animate the character’s mouth and head.

The team, lead by Adobe Researcher Dingzeyu Li, together with Yang Zhou (University of Massachusetts, Amherst) and Jose Echevarria and Eli Schectman (Adobe Research), actually fed their model with thousands of hours of video of real people talking to the camera on YouTube. Surprisingly, that model transferred really well to drawing and paintings — even though the faces the team worked with, including pretty basic drawings of animal faces, don’t really look like human faces.

“Animation is hard and we all know this,” Li told me. “If we all know that if we want to align a face with a given audio track, it is even harder. Adobe Charter Animator already has a feature called ‘compute lip sync’ from scene audio,’ and that shows you what the limitations are.” The existing feature in Character Animator only moves the mouth, while everything else remains static. That’s obviously not a very realistic look. If you look at the examples embedded in this post, you’ll see that the team smartly warps the faces automatically to make them look more realistic — all from a basic JPG image.

Because it does this face warping, Project Sweet Talk doesn’t work all that well on photos. They just wouldn’t look right — and it also means there’s no need to worry about anybody abusing this project for deepfakes. “To generate a realistic-looking deepfake, a lot of training data is needed,” Li told me. “In our case, we only focus on the landmarks, which can be predicted from images — and landmarks are sufficient to animate animations. But in our experiments, we find that landmarks alone are not enough to generate a realistic-looking [animation based on] photos.”

Chances are, Adobe will build this feature into Character Animator in the long run. Li also tells me that building a real-time system — similar to what’s possible in Character Animator today — is high on the team’s priority list.

Read more:


SD Times news digest: ActiveState adds Python packages, Apache Software Foundation on .org registry, and Coralogix joins the cybersecurity market

ActiveState announced that it added more than 50,000 package versions covering the most popular Python 2 and 3 packages to its ActiveState platform. 

“In order to ensure our customers can automatically build all Python packages, even those that contain C code, we’re designing systems to vet the code and metadata for every package in PyPI. Today’s release is a significant first step toward that goal,” said Jeff Rouse, the vice president of product management at ActiveState.

Developers can automatically build open source language runtimes from source, automatically resolve all dependencies, and then certify it against compliance and security criteria within a few minutes. 

Apache Software Foundation speaks out against .org registry
The Apache Software Foundation objected to the for-profit sale of the .org registry, because it believes that a for-profit registry is unlikely to protect the interests of non-profit foundations. 

“This principle of enabling the open exchange of information to and among those we serve lies at the heart of what we do at The Apache Software Foundation,” the Apache Software Foundation wrote in a post.

Coralogix raises $10 million and enters cybersecurity market
Coralogix, provider of an ML-powered log analytics solution announced that it raised $10 million in a Series A funding round, bringing its total to $16.2 million. 

“This new capital will help complete our vision to empower DevOps to control operations and security data from collection through to insights and decisions — with the ultimate goal to make raw log data obsolete,” said Ariel Assaraf, the CEO and co-founder of Coralogix.

The company also announced that it is officially entering the cybersecurity market with the launch of an integrated security information and event management solution (SIEM) and intrusion detection system (IDS) designed for DevOps teams. 

ArcBlock announces 2.0 of its decentralized identity wallet
ArcBlock announced that it is releasing a new version of its decentralized identity wallet that is dynamic and supports users’ daily life. 

ABT Wallet 2.0 now includes a new user-experience for managing accounts and digital properties, easier connectivity and interaction with applications, services and connections and a streamlined way to check activities and transactions. 

The full list of details on the new release are available here. 

The post SD Times news digest: ActiveState adds Python packages, Apache Software Foundation on .org registry, and Coralogix joins the cybersecurity market appeared first on SD Times.

Read more:


Shopping ads come to YouTube’s home feed and search results

There’s a new kind of ad coming to YouTube . Google announced today the launch of Shopping ads on YouTube, which lets brands advertise their products and services right in the YouTube home feed and search results. For example, if a user searches for “Puma shoes review,” a Shopping ad may offer a row of suggested products at the top of the page before the video results.

The ads may also appear as a carousel between the videos on the homepage.

Puma is a debut advertiser for the new Shopping ad product, but the video site will soon fill with these sorts of product suggestions.

“Consumers are continuing to watch more content on the YouTube platform and we want to be where they are, to reach and engage them,” said Rick Almeida, vice president of e-commerce at Puma Group, in a statement about its new YouTube ads. “This new opportunity will enable Puma to extend our shopping strategy into a new property and inspire consumers,” he added.

As Google explains, the ads can be shown to YouTube users based on their interests.

To continue the Puma example, the user wouldn’t necessarily have to type in “Puma” to encounter an ad for the running shoes — simply expressing an interest in running could have them coming across ads from Puma or any other retailers offering running apparel.

Like the Shopping ads that appear elsewhere across Google’s platform — including Search, Shopping, partner websites and the Google Display Network — the YouTube Shopping ads will match to user’s interest not by using keywords but rather on the product details and information the brand submits through the Merchant Center.

The idea to leverage YouTube as a new platform for visual advertising comes at a time when other social networks — like Instagram, Pinterest and even TikTok — are making it easier for users to shop products from their apps. Pinterest has been working to capture shopper interest earlier in the journey, then track the path from visual inspiration and pinning all the way through to purchase.

Instagram this year launched shopping checkout, allowing users to transact from sellers without leaving the Instagram app. More recently, TikTok launched a “Hashtag Challenge Plus” product that lets video viewers shop for products in its app, as well.

But YouTube hadn’t yet fully capitalized on its ability to direct its audience to specific products, rather focusing on Discover ads that would include a visual and a few lines of text, but not necessarily a unique product.

Google says advertisers already using standard Shopping campaigns today and who are opted in to YouTube on Display Network will be immediately able to run YouTube Shopping ads.

The new ads are only one of several changes YouTube announced today. It also said its video ads will now be more interactive, giving users actionable information like store location, interest forms and additional calls-to-actions to help drive more conversions. It’s also rolling out sitelink extensions for TrueView for action ads that will allow viewers to navigate to additional landing pages, like those for holiday catalogs, store hours and more. These will come in the months ahead.

Elsewhere on Google, Showcase Shopping ads are expanding to Google Images, where users will be able to explore a larger selection of products from a brand.

Google had announced its plans to bring new ad products to YouTube back in May, when it revamped the Google Shopping product following the closure and rebranding of Google Express.

As a part of that larger update, the company mentioned a variety of ways it would be connecting the YouTube audience more directly with brands and products — including through its highly visual Showcase Shopping ads and via Shopping Actions, which allow for purchases right from Google’s platforms.

The larger goal with the new ads is to appeal to users with more visual imagery, as today’s web users no longer just search and click on the links that return.

Read more:


Google Maps Incognito mode starts rolling out for Android users

We’ve known for a while now that Google was bringing the “Incognito mode” concept to Maps, allowing you to run searches and find routes without them automatically being tied to your account history.

If you’ve been digging around trying to find the option without any luck, you weren’t just missing it. Though first mentioned back in May at Google I/O, the company says the rollout is just now officially underway.

Word of the rollout comes via a Google Maps support page, as first spotted by AndroidPolice.

It’s a staged rollout, so don’t be surprised if you don’t see the new feature immediately, even if you’re on the latest version of maps. It’s rolling out in batches, beginning with Android users. Google says it should be available to all Android users in “the next few days.”

Once it’s enabled on your account, you can toggle incognito mode on/off by tapping your profile picture, then flipping the switch. Here’s what that looks like:

So why incognito mode? As we wrote back in May: Whether it’s the holiday season and you’re trying to keep your gift-hunting locations under wraps, or you’re visiting a doctor and would just prefer it not pop up the next time a friend grabs your phone for some quick directions, there are all sorts of reasons you might want to leave fewer breadcrumbs. Remember, though, that while it’s less visibly tied to you, it’s still all stored in ways behind the scenes on Google’s end; the company told Wired earlier this month that while Incognito sessions aren’t tied to an account, they are logged with a unique session identifier that gets reset between sessions.

Read more:


SD Times news digest: Amazon announces Rekognition Custom Labs, Scala.js 1.0.0-RC1, and Julia 1.3

Amazon announced Amazon Rekognition Custom Labels, which enables customers to build specialized ML-based image analysis capabilities that detect objects and scenes integral to their specific use case.

“Instead of having to train a model from scratch, which requires specialized machine learning expertise and millions of high-quality labeled images, customers can now use Amazon Rekognition Custom Labels to achieve state-of-the-art performance for their unique image analysis needs,” Amazon wrote in a post.

For example, users can train models to identify specific machine parts such as turbochargers and torque converters after inputting 10 sample images for each specific part to start. 

Scala.js 1.0.0-RC1 released 
Scala.js 1.0.0-RC-1 has been released, which drops support for sbt 0.13x and Scala 2.11.{0-11}, adds tweaks in the JSEnv API and the Scala.js linker is now loaded by reflection in the sbt plugin. 

At the language level, the only major change is that in Scala.js 1.x, ‘’ and ‘@JSGlobalScope’ objects refer to the global scope of JavaScript, rather than the global object. 

The full list of changes in the new release is available here.

Julia 1.3 released
Julia 1.3 has been released with new language features such as support for Unicode 12.1.0, the ability to add methods to an abstract type, support for Unicode bold digits and double-struck digits 0-9 as valid identifiers. Julia is a high-performance, dynamic programming language 

Also, the ‘ClangSA.jl’ static analysis package has been imported, which makes use of the clang static analyzer to validate GC invariants in Julia’s C code. The analysis may be run using ‘make -C src analyzegc.’

The full list of changes are available here.

Instana adds Pivotal Container Service support
Instana added Pivotal Container service (PKS) support to manage a broader range of Kubernetes-based applications with a single management tile. 

“With the ability to achieve full automated visibility with a single click, the Instana tile gives Operations and Development professionals the observability they need to ensure their applications perform at the highest levels,” said Pete Abrams, Instana co-founder and COO. 

PKS users get automatic discovery and monitoring of all Kubernetes nodes, pods and clusters, correlated monitoring and analysis from microservice and end to end tracing of every request.

The post SD Times news digest: Amazon announces Rekognition Custom Labs, Scala.js 1.0.0-RC1, and Julia 1.3 appeared first on SD Times.

Read more:


Manage your angel investors, or they’ll manage you

Lewis Hower

Share on Twitter

Lewis Hower connects SVB and VC/startup communities as a Managing Director with SVB Startup Banking.

In 2016, angel investors pumped approximately $24 billion into startups.

It’s almost certain that an angel will play a role in your startup’s journey, but like everything else in a startup’s life, you need to watch out for potential problems. If you don’t manage them properly, these early backers could get in the way of your startup’s success. The advice from investors and founders below will help you navigate the process of raising a successful angel round while avoiding some of the long-term hassles.

James Currier laughs at how little he knew as a first-time founder looking for angel investors in 1999. The more investors the better, he figured — until he had to handle the fallout of an angel round with sixteen backers.

“They want you to talk to their lawyer and their tax accountant and, before you know it, my 16 angels turned into 64 different relationships,” says Currier, a four-time serial entrepreneur and now a managing partner at early-stage venture firm NFX . “It quite quickly became almost a full-time job checking in with them and taking time to hear all their ideas.”

His advice to founders? Keep your circle of investors as small as possible so you can concentrate on what matters. “This is a time to be as focused as you can on product and customers and revenue,” he says. Raising an angel round is a triumphant moment for any fledgling startup. Yet as with every step of a founder’s journey, there are potential landmines along the way. Failing to manage your angels, Currier and others warn, can distract a company and hurt, rather than help, its chances of success. “They can really gum up the works sometimes,” Currier says.

Be selective

The first thing to keep in mind when thinking about raising an angel round is to choose carefully. “I advise clients to be really thoughtful about who they bring into the seed round,” says Ivan Gaviria, a lawyer at Gunderson Dettmer who has been counseling Silicon Valley startups for more than twenty years. “Clients will say to me, ‘Oh, I need this person because they’re connected in this industry and they’re going to get me leads or whatever,’” he says. “But guess what? Everybody’s well-intentioned, but everybody’s also busy.”

In the vast majority of cases, once angels write a check and get their shares, Gaviria says, “they are not going to spend a ton of time adding value to the company.” Yet Gaviria still sees entrepreneurs pursue what he and others call a “party round,” especially when founders have a well-developed network. “I’ve seen 20, 25, 30 parties, all wanting to drop $25,000 or $50,000 in an entrepreneur’s new endeavor,” he says. “They’re trying to do right by their friends and acquaintances and others who want in on the angel round, but they’re also creating headaches for themselves and their lead investors.”

Gaviria gives the example of “pro-rata rights” — the routine guarantee that angel investors can choose to buy a proportionate number of shares in a future round. “Now 20 or 25 or 30 people have to be notified and the paperwork done for each,” he says.

Set expectations and boundaries for communication

An even bigger challenge is time management and navigating relationships with 20 or 30 people to whom you’ll now feel obliged. “There’s a difference with angels between information versus advice and engagement,” Currier says. “As a CEO, you have to explain to angels the difference. Because some angels want to be entertained.” Those investors hope for a financial return, of course, but they also relish the idea of having a front-row seat to your entrepreneurial journey. To make sure that desire doesn’t turn into a burdensome series of check-ins and inopportune “how is it going” calls, NFX gives founders a template for communications with investors. A monthly report helps to set both expectations and boundaries, Currier says.

Read more:


As sales of smart speakers grow, Soundcheck wants to help make the web more ‘speakable’

Slowly but surely, smart speakers are taking over. As Amazon builds Alexa into everything from tiny clocks to microwaves and Google wraps Assistant into just about anything it can, it feels like it’ll be no time before the rooms that don’t have some sort of voice-powered device are the exception.

But most people and businesses probably have no idea how to get their content prepped and ready to play friendly with these speakers. That’s the driving force behind Soundcheck, a company opening its doors this morning.

Soundcheck helps to take your content and package it in a format these smart speakers and voice assistant devices more readily understand.

Soundcheck’s primary focus initially is on WordPress -powered sites — not a small target, considering that estimates suggest WordPress powers over 30% of the internet. They’ve built a plugin that lets you take information on your WordPress site and, in a tap or two, wrap up the most important pieces in Google’s “speakable” data format — effectively acting as a highlighter, saying, “Hey voice speakers and the search algorithms that power them! This bit of information is meant for you, and answers that question about Topic X.”

soundcheck screen

Getting data into this format usually means writing custom markup for each page in question, which is something that not everyone (like, say, a small business owner using WordPress mostly for the whole simplified WYSIWYG aspect) is prepped to do. Soundcheck boils down the process to a button press, handles the data validation and provides a preview of how that content might sound when read aloud by a voice assistant.

Soundcheck will be free for users who just want the basic plugin, with support for their 50 latest WordPress posts. If you need support for more posts, or you want to do fancier things like custom API integrations and tying into dedicated Amazon Alexa/Google Assistant apps, they’ll charge somewhere between $20-$79 a month. The company tells me it’s also building out an analytics tool to help publishers better understand where and when its data is being accessed by voice. They also say that support for other content platforms beyond WordPress is on the roadmap.

Soundcheck was founded by Daniel Tyreus and Narendra Rocherolle — the latter of whom also co-founded Webshots, the ultra early photo sharing site that sold to Excite@Home for $82.5 million in 1999. The duo originally set out to build Peck, a service founded in 2016 that aimed to figure out the best way to pull in information on a subject and pack it down into its most concise form. They found that one of the toughest parts of that equation was getting data packaged up and ready for smart speakers like Alexa and Google Home — so they pivoted to focus on that.

The team has raised $1.5 million to date, backed by True Ventures, Resolute Ventures, Twitter co-founder Biz Stone and Flickr co-founder Caterina Fake — along with Automattic, the very team behind WordPress.

Read more:


Sentry brings mobile application error monitoring to Android

Sentry has announced a new software development kit and native development kit for Android developers. The new kits aims to bring mobile application error monitoring to the Android operating system. 

According to a recently released report, mobile users are expected to grow to 7 billion by 2021 with annual mobile app downloads projected to reach 258 billion by 2022. However, developers will need more than downloads in order to achieve app success. Without proper tools to help them identify and address errors quickly, developers are at risk of app crashes and damaged brand reputation, the company explained. 

RELATED CONTENT: The coming mobile disruption

“App crashes cause more than 70% of uninstalls, and Google ranking algorithms now downrank apps with stability problems, so uptime and performance are critical for companies to remain successful and competitive in the mobile world,” said David Cramer, co-founder and CEO of Sentry. “The challenge for mobile application developers is the lack of visibility and control over the devices. And on Android, flaws in your native code, third-party dependencies or, in rare cases, even in the system libraries, can bring down the entire application.”

According to the company, the SDK and NDK feature the ability to trace bugs all the way into native libraries, provides information about the phone, and automatically collects the state of the phone. In addition, every crash and report is captured in real time, and developers can attack context information related to the error such as breadcrumbs, tags and other details.

“Errors aren’t platform-agnostic, so why should your error-monitoring services be? Sentry support for mobile, coupled with proven support for web, gives developers a complete picture, which is key in today’s complex application-centric landscape,” said Cramer. “Modern applications are not self-contained—they have multiple runtimes across the stack, causing added complexity in monitoring. While most application error monitoring solutions focus on just one platform, Sentry provides unmatched value across web, mobile and native applications.”

Other features include the ability to see the kinds of phones impacted and specific actions a user took, and an API to add application-specific data and events to crash reports. Sentry for Mobile is also available for iOS, React Mobile, Flutter and other mobile platforms.

The post Sentry brings mobile application error monitoring to Android appeared first on SD Times.

Read more:


A network of ‘camgirl’ sites exposed millions of users and sex workers

A number of popular “camgirl” sites have exposed millions of sex workers and users after the company running the sites left the back-end database unprotected.

The sites, run by Barcelona-based VTS Media, include,, and Most of the sites’ users are based in Spain and Europe, but we found evidence of users across the world, including the United States.

According to Alexa traffic rankings, is one of the most popular in Spain.

The database, containing months-worth of daily logs of the site activities, was left without a password for weeks. Those logs included detailed records of when users logged in — including usernames and sometimes their user-agents and IP addresses, which can be used to identify users. The logs also included users’ private chat messages with other users, as well as promotional emails they were receiving from the various sites. The logs even included failed login attempts, storing usernames and passwords in plaintext. We did not test the credentials as doing so would be unlawful.

The exposed data also revealed which videos users were watching and renting, exposing kinks and private sexual preferences.

In all, the logs were detailed enough to see which users were logging in, from where, and often their email addresses or other identifiable information — which in some cases we could match to real-world identities.

Not only were users affected, the “camgirls” — who broadcast sexual content to viewers — also had some of their account information exposed.

The database was shut off last week, allowing us to publish our findings.

The “camgirl” site, which exposed millions of users’ and sex workers’ account data by failing to protect a backend database with a password. (Image: TechCrunch)

Researchers at Condition:Black, a cybersecurity and internet freedom firm, discovered the exposed database.

“This was a serious failure from a technical and compliance perspective,” said John Wethington, founder of Condition:Black. “After reviewing the sites’ data privacy policy and terms and conditions, it’s clear that users likely had no idea that their activities being monitored to this level of detail.”

“Users should always take into consideration the implications of their data leaking but especially where the implications could be life altering,” he said.

Data exposures — where companies inadvertently leave their own systems open for anyone to access — have become increasingly common in recent years. Dating sites are among those with some of the most sensitive data. Earlier this year, a group dating site 3Fun exposed over a million users’ data, allowing researchers to view users’ real-time locations without permission. These security lapses can be extremely damaging to their users, exposing private sexual encounters and preferences known only to the users themselves. The fallout following the 2016 hack of affair-focused site Ashley Madison resulted in families breaking up and several reports of suicides connected to the breach.

An email to VTS Media bounced over the weekend and could not be reached for comment.

Given both the company and its servers are located in Europe, the exposure of sexual preferences would fall under the “special categories” of GDPR rules, which require more protections. Companies can be fined up to 4% of their annual turnover for GDPR violations.

A spokesperson for the Spanish data protection authority (AEPD) did not respond to a request for comment outside business hours.

Got a tip? You can send tips securely over Signal and WhatsApp to +1 646-755-8849. You can also send PGP email with the fingerprint: 4D0E 92F2 E36A EC51 DAAE 5D97 CB8C 15FA EB6C EEA5.

Stop saying, ‘We take your privacy and security seriously’

Read more:


Workplace learning platform HowNow scores $3M funding

HowNow, the workforce learning platform, has raised $3 million (£2.4m) in a “pre-series A” funding round. The round is led by Mark Pearson’s Fuel Ventures and brings the total raised by the startup to $4.5 million.

Other investors include Andy Murray OBE; Michael Whitfield and Chris Bruce (founders of Thomsons Online Benefits); Bernie Sinniah (former managing director at Citi Bank); and Alwin Magimay (a former partner at McKinsey).

Designed for organisations that want to support teams with self-directed learning and the development of “business-critical” skills, HowNow is described as an integrated learning platform that autonomously curates learning resources, “business intelligence” and market insights that live in various internal and external sources.

The idea is to bring together these different learning resources — ranging from “nuggets” of knowledge shared by existing employees to internal data to external content libraries, blogs, podcasts — and match these to different job descriptions and employee skill-sets.

This is powered by a browser extension and integrations with Slack, Salesforce, Hubspot and over 300 other apps. Machine-learning is also employed to push the right content to the right employee.

“Employers can also use HowNow to identify skills gaps within the company based on job market data, via HowNow’s real-time analytics and built-in certification,” adds the company. To achieve this, the platform claims to monitor over 20,000 job specifications to understand the in-demand skills and requirements companies are searching for.

“Based on self-review, peer-review and real-time job market data we build the user’s skill profile as they onboard the platform,” explains HowNow co-founder and CEO Nelson Sivalingam. “Once in HowNow, they see learning recommendations based on assigned learning pathways, their role, skill requirements and internal benchmarks. This content is brought together from a variety of their internal sources (G Drive, Sharepoint, CRM, etc), external sources (content libraries, blogs, podcasts, etc) and the autonomously organised knowledge shared by their peers directly on HowNow”.

Employees can then access these learning resources directly within the applications they already work with and receive contextually relevant suggestions powered by HowNow’s “AI”. “For example, they can be in Slack and search all of their learning resources directly from their using the HowNow Slack app,” says Sivalingam. “They can also convert a message from a colleague into a nugget that will get stored and autonomously organised in HowNow”.

Similarly, Sivalingam says that, via HowNow, client facing teams are able to access up-to-date product knowledge, business intelligence and market insights directly within their inbox, CRM and helpdesk, which enables them to reduce customer response times.

“Fast-growing companies like GymShark are able to capture the knowledge in the heads of their internal subject matter experts by giving them a quick and easy way to share knowledge, build a glue between scattered content, avoid repeat questions and get everyone on the same page,” he adds.

To that end, I’m told that more than 500,000 users currently use HowNow within over 125 businesses. These range from SMEs to larger organisations, across 14 different countries. A classic SaaS play, the startup generates revenue through a licence fee per user.

Read more: