Project Snowflake
Leon Silverman, head of Disney’s Digital Studio, remarked that every production is like a snowflake – each one is unique. Media production workflows are complex and in flux. The choices filmmakers must make to create their project seem infinite and daunting – Which camera? What workflow is best, given the size and scope? Which processes work best in post-production? Studios can afford to fully test multiple cameras and workflows, but DIY and Indie filmmakers cannot. Funded by Google Research, the Snowflake project was launched to replace this testing process with data-driven color and emotion analyses, and a database of tried and tested workflows.
- Lookbooks play a fundamental role in the process of media ideation and creation. They convey mood, tone, emotion and visual style to team members and financial stakeholders. Snowflake is an expanded lookbook creation tool that centers on transmedia production workflows in a collaborative environment using machine learning.
- Although a range of data-driven solutions support media making, the vast majority of these tools rely solely on text and numerical data as their input. Snowflake takes a different approach to the collaboration between human and ML/AI participants, positing algorithmic support for the creative process. This approach resembles the processes and ideas of groups like Oulipo or the Surrealists, rather than big data processing. Similarly, the Snowflake project endeavors to integrate algorithmic processes as the foundation of inspiration fostering creative collaboration among all parties involved. Machine learning, in this context, is designed to support innovation by relying on a cinematically specialized visual database.
- In the intuitive process of the creation of a lookbook, “Flakey,” the ML agent offers suggestions based on an analysis of creative input (image, sound, text), cross-referencing it with an underlying database. This kind of human-machine collaboration supports an interaction that is beyond a language-based search, using computer vision and sentiment analysis algorithms to offer a more intuitive, associative, and serendipitous way of conveying emotionality and tone in a form of a lookbook.
Project Site: https://snowflake.usc.edu/
Project Team:
- Scott Fisher
- Eric Furie
- Norm Hollyn
- Andreas Kratky
- Virginia Kuhn
- Mediatrix Lopez
- Szilvia Ruscev
With special help from:
- Maduabuchukwu Udeh, Summer Xiang, Katrina Xiao, Cecile Zhang
Project Sponsor:
- Google Focused Award Program: The Future of Storytelling