Artbreeder is an art website where users can generate and modify images of faces, landscapes, and paintings, among other categories. The tool uses machine learning to create these works in a wide range of styles, from impressionism to surrealism.
The site was founded in March 2018 by an engineer named Joel Simon as Ganbreeder; it was later rebranded to its current name in September 2018 following a major update that included new features. According to Joel, “It is also inspired by an earlier project of mine Facebook Graffiti which demonstrated the creative capacity of crowds.”
How does Artbreeder work
The tool runs on three key components: an image generator model, an image discriminator model, and a loss function that defines how well the generator did at imitating real data. The model uses two competing neural networks that work together in order to generate original pieces of artwork based on pre-trained models using different content categories. First, one neural network is responsible for creating these pieces while another is in charge of determining whether or not the image is real or fake. As this method relies on machine learning, the content categories are defined by classes, which are created using training data using images from different sources.
To prevent overfitting and ensure good quality output, the user can set selection criteria for both generator and discriminator models to generate high-quality results with desirable properties. The generator model is trained with thousands of images that were initially used to train it. When generating new images based on inputted content, Artbreeder applies an additional multi-scale transformation composed of global average pooling followed by local deformations to its inputted images while also resizing them.
The process begins when a user uploads their preferred content onto the platform. They will be provided with a link to share their design and can share this link through social media or email. This content is then fed into the discriminator model which sits between the generator and inputted images, allowing it to differentiate between real and fake or generated images based on its training phase. The user can select from different GAN models depending on what kind of results they want from the image generation process.
The output is a high-quality image that has been auto-generated using the selected GAN models, but this does not mean that users have no control over what kinds of images are created. If a user uploads an image that may result in undesirable scenes being generated, such as a landscape with trees lacking leaves during wintertime, there is an option to disable the generation of images that are likely not desired. By doing so, the user will have more control over how content they upload can be used when generating new images.
The process is then repeated with thousands of other images provided in the training dataset that are used to generate new results by altering existing inputted content through a number of different factors. While some users may want to explore this creative process in its entirety, others might prefer to start by having fun or exploring specific genres, which can all be done on Artbreeder using machine learning models.
Why should you use it?
If you are an artist, you can experiment with machine learning and see how it works. You could also use Artbreeder to train on existing paintings before starting your own painting process. For business owners, using the tool helps in marketing through art creation. It helps in building artistic profiles for brands by creating branded artwork that is unique to them.
Is Artbreeder free?
Another reason why you should use it is that Artbreeder is available for free on the website. When uploading content onto the platform, users are treated to different GAN models that can be selected depending on what kind of results they want to achieve by altering their preferred images.
Artbreeder is a user-friendly platform that can be used by artists both as a tool for creation and as an exploration of the creative process.