Photolab is a collection of experimental apps, concepts, and prototypes created to explore Google AI machine perception research, and improve the feedback cycle between Google researchers, users, and products. We aimed to maximize learning with small experiments, and to find appropriate fits for AI research with relevant and accessible user experiences.
︎︎︎Zoop (below): a mini-game that explores challenges of reinforcement learning. Reinforcement learning (RL) encourages machines to determine ideal behavior based on a reward system. Typically, a human needs to manage the system and train the machine in learning appropriate behaviors. Zoop breaks down training into one task per level (moving forward, jumping, etc.) to illustrate how a machine learns during RL.

︎︎︎ A Story A Day (below): creates one children's story per day. A machine-learning model trained on children's stories would power the experience, plotting a storyline and providing users with opportunities to guide the story
by occasionally choosing the ends of sentences, key verbs, or nouns. The illustrations are Quick, Draw! entries from their collection of over 50 million drawings.

In December 2017, our 9-person team at Google AI launched three experimental apps to share and gauge public interest in our work.

Selfissimo [iOS] [Android] explores automating the selfie experience, or relinquishing some control back to the "photographer" behind the camera.
Storyboard transforms users' videos into comics. It selects, composes, and stylizes video frames using experimental ML style transfer technology that runs locally.

allows users to manipulate the speed and direction of video playback to create non-linear, creative remixes.

︎︎︎Concepts and ideas: In addition to our more public-facing work, we also generated hundreds of ideas around the future of photography, from the purely imaginative to those grounded in visual perception technology from Google AI.
Below are a few examples of early concepts we visualized, including "interactive photos" and a new perspective on slit-scan photography.