What is artline?

ArtLine is a machine learning project that converts portrait photos into line art drawings. You give it a photo of a person's face, and it outputs a black-and-white line drawing in the style of a hand-drawn portrait sketch. The project is focused specifically on faces and portraits rather than general images. The simplest way to try it is through a Google Colab notebook, which runs in a browser without any installation. Two versions are available: a faster model that produces smoother results and a slower one that aims for higher quality. There is also a newer mode that combines ArtLine with a tool called ControlNet, where you can provide both a photo and a written instruction to influence the style of the output. The creator trained the model by combining two datasets: one containing portrait photos paired with hand-drawn sketches, and one from anime sketch colorization. The combination helped the model learn to handle a wider variety of poses rather than only straight-on face photos. The technical approach uses a neural network architecture designed for image-to-image tasks, with techniques borrowed from a related project called DeOldify. The model has acknowledged limitations: it works best with well-lit, high-quality photos against clean backgrounds, and it tends to confuse shadows with hair. Images below 500 pixels wide often produce poor results. The project was created by someone who describes themselves as not primarily a coder, and it is an ongoing personal project that has been updated over time. Source code is under the MIT license.