Sneaks—those quick peeks at still-in-development-technology from Adobe — are always one of the highlights of MAX, and 2021 did not disappoint.
During the all-virtual conference, presenters from Adobe Research revealed some of the most exciting new experimental creative tools on the horizon, with help from Saturday Night Live’s Kenan Thompson.
Among the Sneaks were new tools that create and animate images between similar photographs, search for photos with people in any pose, tweak portraits to create just the right pose, add realistic shadows, build customized fonts, and more.
Here’s a look at the seven Sneaks from Adobe Research this year.
Project In-Between uses the power of Adobe Sensei to generate an animated bridge from two or more similar pictures. Users can pause the animated imagery to view and choose among the dozens of newly generated in-between frames created by the tool’s AI. The “living photos” that users create with this experimental technology are perfect for sharing on social media as GIFs.
Presenter: Simon Niklaus
Collaborator: Oliver Wang
Project Shadow Drop
Traditional shadow rendering methods can be tricky, as they require geometric knowledge and a familiarity with lighting sources. Project Shadow Drop solves this problem by using the 2D position of a light source and the horizon and allowing you to automatically generate realistic shadows which can be applied to 2D vector art, 2D animations, and even real images. For designers, artists, and animators alike, this can enable shorter turnarounds by eliminating the need for time-consuming, granular edits.
Presenter: Jianming Zhang
Collaborators: He Zhang, Zhe Lin, Eli Shechtman, Yichen Sheng, Yifan Liu
Project Sunshine takes vector graphics to the next level, providing automated suggestions for coloring and shading options. The generative model behind Project Sunshine is auto-regressive, meaning it starts by guessing an element of the image (e.g. “the hair should be black”) and then spirals outward from this decision. The autoregressive nature of the model makes it easy to generate many diverse examples. Best of all, because the results are vectorized, it’s easy to continue editing and refining the color and shading suggestions.
Presenter: Matthew Fisher
Collaborators: Deepali Aneja, Praveen Kumar Dhanuka, Vineet Batra, Ankit Phogat, Sumit Dhingra, Nitin Sharma
Project Artful Frames
Project Artful Frames aims to simplify the time-intensive process of creating artistic animations. Artful Frames combines AI-learned representations, optimization, and super-resolution to propagate the style from a single example of an artist’s work to a live video of a completely different scene. The artwork provides the artistic direction, while the video provides the layout and realistic motion. Artful Frames could one day help artists rapidly explore different options and even open up new creative possibilities that would otherwise be too labor intensive.
Presenter: Nick Kolkin
Contributors: Eli Shechtman, Sylvain Paris
Project On Point
Finding the right stock photo can seem like an endless task, especially if you’re looking for a model in just the right pose. Project On Point improves image search with pose-based descriptors. The descriptors are interactive — they’re represented as a 2D stick figure layered over your referenced image — and can be modified for further query refinement. The descriptors can also be easily combined with other search modalities, filtering for images tagged with specific attributes such as “man,” “woman,” or “child.” Project On Point isn’t limited to stock photo databases. You can also use it within your own photo albums, making it an ideal option for searching a recent fashion shoot or portrait session.
Presenter: Duygu Ceylan
Collaborators: Jimei Yang, Jun Saito, Jinrong Xie, Shabnam Ghadar, Baldo Feita, Alex Flipkowski
Project Stylish Strokes
Fonts can convey personality, tone, and creativity, but they often leave much to be desired in terms of customization. Fonts are typically stored as outlines, so stylization or animation are difficult to achieve because the edits must be based on the underlying strokes that make up each individual character within a font. Project Stylish Strokes automatically recovers those strokes based on the geometry of the characters, enabling a wide variety of stylizations that work for fonts in any language, and even for unusual character structures. Project Stylish Strokes makes the options for colors, textures, and animations of fonts endless.
Presenter: Paul Asente
Collaborators: Jose Echevarria, Daniel Berio
Project Strike a Pose
Posing for a photo can be awkward. What exactly are you supposed to do with your hands? Project Strike a Pose solves the problem by allowing you to go back to an image and adjust a model’s pose. You provide a reference image of a person in a desired pose, then Project Strike a Pose leverages machine learning to reposition the person in your image into the same stance. Project Strike a Pose is able to replicate elements such as clothing, hair, and skin color to match the source image while retaining the desired pose from the reference image.
Presenter: Krishna Kumar Singh
Collaborators: Tuanfeng Wang, Cynthia Lu, Duygu Ceylan, Yi Zhou, Giorgio Gori, Xin Sun, Jimei Yang, Yijun Li, Tobias Hinz, Yang Zhou, Ruben Villegas, Nathan Carr, Niloy Mitra, Hailin Jin, Celso Gomes, Eli Shechtman, Scott Cohen, Jun Saito, Zhixin Shu, Cameron Smith, Daichi Ito, Badour Albahar