Playtime with Bernie
⌝ Mixed-media generative AI, ML & Unreal Engine realtime ⌞
How can we achieve innovation through learning and experimentation, by embracing a beginners mind? How can we meaningfully introduce machine learning and AI tools into our creative process?
Bernie’s world is a playful research & learning project which translates physical drawing into a complex digital shape-system with the help of various new AI tools and processes.
Using a series of hand-drawn shapes translated and expanded into data sets, a series of characters are created to inhabit a playable world full of unique forms.
For the first edition of OPN Motion Festival in London, the FutureDeluxe team devised a framework to answer these questions. Instead of presenting past work, we decided to experiment through play and show the audience of mostly students our approach to work as play. All games have rules. Here are ours: — Work with someone you've not worked with before — Work with AI — Learn a completely new tool.
The project found inspiration in one of the London studio team member's hand-drawn doodles. The sketches were digitised to find out if we could train a custom ML model on the doodles and have the computer generate new ones.
The input images are processed to create large custom datasets.
Experimental learning through doing is a a playful process designed to shift creative attention away from result-driven performance and back into an exploratory creative space of gradual progression. Allowing space for experimentation and new tools is essential.
A custom model was trained on the digitised datasets of the doodles in order to learn the shape language. The outputs were then further manipulated and used as creative raw material.
We then scatter the shapes across a basic character rig in order to begin to generate characters forms ↓
The system allows us to generate unexpected options. Parametrically scattered shaped were rendered from the Houdini viewport, which was then used as a guide for the custom-trained model to generate multiple characters.
After completing the character design, the team selected "Bernie" and built a custom real-time world for it. Bernie's world consists of both available and custom-design assets from the same dataset as Bernie itself.
Outcome: The team designed a custom pipeline and learning experience which focused not only on the technical aspects but on the artistic challenge that new tools integration requires.
Thomas O’Reilly
Yas Vicente
Margot Parker
Svet Lapcheva