Google AI’s Chimera Painter is a demo tool that uses machine learning (ML) to enable users to create a fleshed-out creature based on a rough sketch. Chimera Painter adds features and textures to a creature outline, segmented with body part labels such as claws or horns. The ML model is created with Unreal Engine and the GAN-based (generative adversarial network) ML algorithm. Although the tool is primarily aimed at game developers, it can be used by anyone to turn a rough doodle into a realistic-looking monster, just for fun.
“What if artists had a paintbrush that acted less like a tool and more like an assistant? A machine learning model acting as such a paintbrush could reduce the amount of time necessary to create high-quality art without sacrificing artistic choices, perhaps even enhancing creativity,” Google AI team wrote in a blog post, explaining the Chimera Painter demo in detail.
Chimera Painter has four preset creature outlines to choose from. These can be modified using the tools available on the left of the screen, of 20 different body parts such as head, horns, teeth, claws, wings, tails, and more. Each has a specific colour assigned that will help you differentiate between body parts when you’re drawing. You can adjust the size of the paintbrush and use the eraser and clear tools as well.
After you’re done with the creation, click on Transform to turn your doodle into a realistic-looking creature. You can also use the tools to add any part anywhere – you could add horns on your monster’s torso and teeth to its tail, and the tool will do its best to incorporate that.
You can also upload a creature outline created in an external program like Photoshop. Download one of the preset creature outlines to get the required colours needed for each creature part and use it as a template. Then, click on the Load button on the demo to use that outline.
Google said that to pursue this goal, the team created a fantasy digital card game prototype around the concept of combining creatures into new hybrids that can then battle each other. To train the generative adversarial networks (GANs), they created a dataset of full colour images with single-species creature outlines that were adapted from 3D creature models. After model training, it was tasked with generating multi-species chimeras that were based on outlines provided by artists.
The best-performing model was then incorporated into Chimera Painter, explained the Google AI team. A new artist-led, semi-automated approach was also developed to create an ML training dataset from 3D creature models.