Google developers have introduced a new tool – Chimera Painter. It uses machine learning to create images based on user sketches. According to the development team, this is just a prototype, but if such software is improved, it can shorten the time it takes to create high-quality content.
Making art for digital video games requires a high degree of artistic creativity and technical expertise. Plus, gamer artists need to integrate ideas quickly and create massive amounts of data, often under tight deadlines. Google AI thought, what if artists had a brush that acted not as a tool, but as an assistant? A machine learning model acting as such a brush will reduce the amount of time it takes to create high-quality art.
And today, Google engineers unveiled the Chimera Painter tool, a trained machine learning model. It automatically creates a fully expanded render from the user-supplied outline of the drawn creature.
The researchers initially set themselves the task of creating an illustration for a fictional fantasy card game in which players combine the traits of different monsters and fight them. They trained a machine learning model on a database of over 10,000 monster samples, which were themselves partially generated using 3D models rendered in the Unreal Engine.
Each image is accompanied by a “segmentation map” – an overlay that divides the monsters into anatomical parts such as claws, snouts, legs, and so on. Below is an example of using the demo with one of the preset creature outlines.
After training the model on this data, users can draw their own segmentation map, which will then be rendered using photorealistic textures. If you download Chimera Painter, you will see some pre-installed monsters. However, drawing one of them takes more time and effort than you think.