GAN based generation from brushstrokes of an human artist. Mentioned artist is Özgür Balli, PhD from Hacettepe University Fine Arts. The software consist a basic work that learns many brushstroke samples of the artist to generate its own (Fig.1). The resulting images are representative composition (Fig.3-4). The core algorithm is an artist-critic type generative adversarial neural network (GAN).
Paper This work is mentioned in Ballı, Ö. (2020). TRANSHÜMANİZM BAĞLAMINDA BİR YAPAY ZEKÂ SANATÇI UYGULAMASI: OBv2.Tykhe Sanat ve Tasarım Dergisi, 5 (9), 141-162.
Fig. 1: Generated brushstrokes
This is a Python3 projects using modules:
- OpenCV v4.3.0
- Tensorflow v2.1
- NumPy
- glob
- IPython
- Matplotlib
For GUI, PySide2 and win32api are needed.
The interface is for simply composing generated brushstrokes onto a canvas. At this step, GAN has already trained and custom strokes are obtained. The user may select base strokes or let the software select them randomly.
Fig. 2: User interface
Fig. 3-4: Resulting composition