Chimera Painter: Google’s new AI-powered tool to create fantasy creatures

Published by Pye.ai

Google’s latest AI  web-based tool is all set to tickle your creative side and perhaps awaken the monstrous side of it too. Chimera Painter – Google’s trained machine learning (ML) model helps artists create monsters.

Chimera aims to aid art creation for digital video games. The ML model creates full-fleshed fantasy creatures out of our messy drawings. You can add features, textures and choose from several body parts such as wings, claws, and teeth to create a monster.  Once you have given your inputs, click on ‘transform’ to see your imagination take shape.

According to Google, Chimera Painter is like a paintbrush to an artist. It is less like a tool and more like an assistant.

The working prototype

For the working prototype, Google made a digital card game around the concept of creating hybrid creatures and battling them out against each other. A player could begin with a card with real animals and make them powerful by combing them. This method proved useful in designing a large number of combinations of creatures in a short time.

To create high-quality images, Chimera uses generative adversarial networks (GANs).GAN is a machine learning class that automatically involves learning patterns in input data to produce output that will resemble the input data set.

For Ex: GANs can create a human image from human photographs even though the output image does not belong to any real person. It is an exciting and rapidly growing field in its ability to create real examples.

Images created by GAN

Taken from  Progressive Growing of GANs for Improved Quality, Stability, and Variation.

Training the model

Google created a dataset of full-color images with a single species outline adapted from 3D creature models to train GANs. After preparing the model, the model was tasked with generating multi-species creatures based on the artist’s outline.

Generating creatures using GAN had the potential to lose essential features due to the low contrast parts of images. The model needed a non -photographic fantasy-styled data set. Google solved the problem by creating an artists led semi-automated data set from 3D models.

They trained the model on a database of more than 10,000 sample monsters. Each picture is paired with a segmentation map. A segmentation map consists of flat colors for each body part (e.g., head, ears, neck, etc.)

Photo Credits: Google

Google used Unreal Engine to place the 3D models in a simple 3D scene. A set of scripts would then take this 3D scene and introduce different poses, viewpoints, and zoom levels to create new 3D creature models, creating full-color images. These full-color images and segmentation maps formed the training data set for GAN.

Chimera painter demo is now available for artists. It aims at creating numerous fantasy creatures in reduced time. You can easily adjust the shape, type, and add body parts, allowing you to expand and explore your creative ideas. Chimera Painter also allows for uploading creature outlines from an external program like Photoshop.

I’m no artist, but I gave Chimera a try, and this is what I created. That’s definitely not a beast.

How deadly do your monsters look? Haven’t you tried Chimera yet? Come on, give it a try!

References

Published by Pye.ai

Spread the word

Related posts