FireEye data scientist Philip Tully showed off a convincing deepfake of Tom Hanks he built with less than $100 and open-source code. Until recently, most deepfakes have been low quality and pretty easy to spot. FireEye demonstrated that now, even those with little AI expertise can use published AI code and a bit of fine-tuning to create much more convincing results. But many experts believe deepfake text is a bigger threat, as the GPT-3 autoregressive language model can produce text that is difficult to distinguish from that written by humans. Continue reading Quality of Deepfakes and Textfakes Increase Potential Impact

The Entertainment Technology Center (ETC@USC) VFX Working Group has published a specification for best practices naming image sequences such as plates and comps. File naming is an essential tool for organizing the multitude of frames that are inputs and outputs from the VFX process. Prior to the publication of this specification, each organization had its own naming scheme, requiring custom processes for each partner, which often resulted in confusion and miscommunication. ETC’s specification, which aims to standardize the process for media production, is available online for anyone to use. Continue reading ETC Publishes Specification for Naming VFX Image Sequences

Instagram users have embedded images in their posts, believing that they were protected against copyright claims. Facebook now explains that, “while our terms allow us to grant a sub-license, we do not grant one for our embeds API.” In other words, a user who embeds someone’s Instagram post on her website has to ask the poster in advance for a separate license to the post’s images. Those who don’t could be subject to a lawsuit. Professional photographers will be able to better negotiate with publishers based on these terms. Continue reading Instagram Users Can Be Sued for Embedding Images in Posts

Facebook’s Instagram recently debuted Co-Watching, a feature that makes it possible for users to video chat while they browse the app. With Co-Watching, users can engage in direct-message video-chat conversations to look at saved, liked and suggested photos and videos together. Co-Watching is just one example of ways that Facebook and Instagram are shifting to accommodate users and communities during the coronavirus pandemic, when many are confined to home. Instagram first began testing Co-Watching a year ago. Continue reading Instagram Unveils Co-Watching Feature for Group Video Chat

Twitter announced yesterday that it would be more assertive in identifying fake and manipulated content on its platform. Beginning next month, the company plans to add labels or remove tweets that feature such manipulated images and video content. While short of an outright ban, the new policy is meant to address the growing concern of users frustrated by the practice of disinformation spread via social platforms. However, it also highlights the challenges faced by social media companies in regards to balancing freedom of speech, parody and satire, and false or manipulated content. On Monday, YouTube announced its plans to better manage misleading political content on its site. Continue reading New Twitter Policy Aims to Combat Fake Photos and Video

Clearview AI has scraped over three billion photos from Facebook, YouTube, Twitter, Venmo and hundreds of other websites to create a facial recognition database now used by the FBI, the Department of Homeland Security and over 600 law enforcement agencies. Twitter just sent a cease-and-decease letter to the startup demanding that it stop taking photos and other data from its site and to delete any data it has already collected. Twitter accused Clearview AI of violating its policies. Continue reading Twitter Warns Clearview to Stop Scraping its Users’ Photos

Wooptix approaches the problem of light-field image capture in an interesting way. Using a single camera lens, they capture the same image at multiple depth-of-field settings. They then assemble a 3D version of the captured image by identifying and grabbing the areas of each depth-of-field image that are in focus. The result is a clear, if slightly flat and layered, 3D image with some degree of parallax. The company’s tech could have multiple applications. Continue reading CES 2020: Wooptix Creates 3D Images From a Single Lens

With the beta of iOS 13.2 arriving with the new iPhone 11, Apple is previewing Deep Fusion, the company’s name for machine learning-aided computational photography to enhance details. The trick of snapping multiple exposures that are then composited into an ideal photo might not sit well with professional photographers, but the A13 Bionic-enabled iPhone 11 will do just that, beginning to shoot before the shutter button is pressed and then picking the best bits of the photos to create the best one possible. Continue reading iPhone 11’s Deep Fusion Creates Composited HDR Images

Pinterest is now promoting itself as a “visual discovery engine,” where its 300 million global monthly active users can not only browse billions of images but also purchase the items they find there. For most of its users, Pinterest is a go-to place for home décor, gardening and personal style, allowing users to “pin” or post photos to create inspiration boards. According to Pinterest, its AI technology can accurately pinpoint 2.5+ billion objects in photos, millions of which can be purchased by clicking on the item. Continue reading Pinterest Becomes a Visual Discovery Engine for Shopping

Recent reports from influencer marketing services indicate that brands and marketers are concerned with the growing costs of working with social influencers. According to a Mediakix survey, more than one-third of marketers in the U.S. explained that the rising cost in this space has become a significant challenge to marketing. A report from Klear points out that nano-influencers on YouTube (with 500 to 5,000 followers) earn an average of $315 per video, and power-influencers (with 30,000 to 500,000 followers) charge an average of $782 per video. Continue reading Costs of Teaming with Social Media Influencers on the Rise

Facebook’s Instagram began testing a new approach with users in Canada two months ago and this week expanded its efforts to include users in Australia, Brazil, Ireland, Italy, Japan and New Zealand. The experiment removes emphasis on the “Like” feature to minimize the pressure to compete, while hopefully creating a more personal and enjoyable experience. Users are still able to see who liked other people’s posts or watched their videos, but there is no longer a running tally of the number of likes and views (however, users can still privately see the counts for their own posts). Continue reading Instagram Expands Tests of Hiding Likes to Reduce Anxiety

Amazon, which launched its new StyleSnap feature to select iOS and Android users in April, will soon make the in-app tool widely available, said company worldwide consumer head Jeff Wilke at the company’s re:MARS AI conference in Las Vegas. Users can reach StyleSnap via a shortcut found by tapping the camera icon in the Amazon app’s upper right-hand corner. Based on image recognition, the machine learning-enabled StyleSnap (and Pinterest Lens competitor) will offer similar items to any photo or screenshot uploaded by a user. The algorithms also incorporate computer vision and deep learning. Continue reading Amazon’s AI-Enabled StyleSnap Is Ideal for Fashion Market

For a long while, Google’s AI research division has been on a mission to grow capable of accurate instance-level landmark recognition and retrieving images — both widely recognized as extremely difficult. To that effort, last year, it released its Google-Landmarks dataset and hosted two competitions for more than 500 machine learning researchers. Last week, it went a step further with the release of its open-sourced Google-Landmarks-v2, which contains double the photos featuring seven times the landmarks.

Continue reading New Google-Landmarks-v2 Dataset Means More AI Training

Google unveiled the Deep Planning Network (PlaNet) agent, created in collaboration with DeepMind, to provide reinforcement learning via images. Reinforcement learning uses rewards to improve AI agents’ decision-making. Whereas model-free techniques work by getting agents to predict actions from observations, agents created with model-based reinforcement learning come up with a general model of the environment leveraged for decision-making. In unfamiliar surroundings, however, agents must create rules from experience. Continue reading Google Debuts Deep Planning Network Agent with DeepMind

As it continues to ramp up its profile as a media company and aims to better compete with popular social platforms such as Snapchat and Instagram, Twitter has been testing a visual sharing tool described in its code as the “News Camera.” While Twitter users can already share images and video, this enhanced camera tool would encourage them to do so more frequently. The feature is accessible via a swipe on the home screen and offers users the ability to add captions to photos, video and live broadcasts before such content goes to the timeline. Continue reading Twitter Is Working on Snapchat-Like ‘News Camera’ Feature