Derivative Artist Bot

What is the Derivative Artist?

Derivative artist is a computer program that learns from other artists and then creates art of its own in cooperation with the user. Computers today are commonly used in the production of visual media, but normally only at the level of a drawing tool under the direct control of a person. Derivative Artist (DA) is the result of a project to add a synthesis component to an ALISA engine, allowing DA to study the works of other artists (or any set of images), and then create a new “painting” derived from a single image (usually a photograph) selected by the user. This process imitates a typical training and creative process of human artists: a period of study of the works of established artists and the world around them, and then the creation of paintings, each stimulated by the presence or memory of a visual stimulus. The user affects the artistic temperament of DA by selecting the set of training images for DA. Once trained, a few parameters of DA may be varied to guide its creative style from abstract to realistic.

How does Derivative Artist work (briefly)?

Derivative Artist paints images pixel by pixel. The process starts with a source image, which is akin to the 'subject' a human artist would work from. Then Derivative Artist consults a knowledge-base created during training by studying a group of images one by one. The images are sliced into millions of pieces and stored. When Derivative Artist is creating new work in the style of a particular artist, it breaks the source image into millions of pieces and searches its training for pieces that are similar. The pieces that it finds are used to make decisions about what to paint on the generated image.

What applications are there for Derivative Artist?

The most obvious use is to employ Derivative Artist in the creation of still image art, as a tool in the artist's toolbox. This includes uses such as advertising, creating artwork for your wall, and ornamentation. Derivative Artist can also be used to create digital animation when it is applied to each frame in a movie or animation. Derivative Artist may also be useful as a tool to apply a 'painted' look to photographs; the filters commonly used for this purpose can look uniform or obviously synthetic, and the results from Derivative Artist may be preferable.

It says that the picture was painted after training on Picasso, but the result doesn't look anything like a Picasso!

Derivative Artist trains by studying another artist's work, but Derivative Art cannot mimic every facet of the original artist's "style". In general, one might say that a style includes elements such as subject matter, framing of the scene, media, palette, texture, and mood. Derivative Artist processes whatever it is given, without regard to the subject matter or the framing of the scene, and these elements may not be stylistically consistent with the original artist. The palette and the texture of Derivative Art are sometimes recognizable as similar to the original artist. The objective of the project is to teach a computer to make art, not to exactly copy any particular artist's style. So Derivative Art produced after training on Picasso will not probably be mistakable for the genuine article. It is true that the work of Picasso is involved in the creation of the result.

Can I use Derivative Artist to make my own art?

Not at this time. Derivative Artist is very CPU intensive, and we do not have the hardware resources to service the Internet community to the standard we would like. It takes 5-15 minutes to create a picture with Derivative Artist using up to date hardware, which means that we would require several high powered servers to handle even a few interested surfers. We are looking for a suitable way to make Derivative Artist available for you to experiment with. That being said, we welcome ideas and requests for images.