iGAN

Why we included this project

iGAN treats a generative adversarial network as something you can steer by hand instead of just sampling at random. Sketch a rough shape or lay down a few colored strokes, and the system searches the latent space of a pre-trained DCGAN, returning photo-realistic images that match your edits in real time. That makes it a useful tool for designers and researchers who want to prototype visual ideas by painting directly on a canvas, and for anyone trying to get a feel for what a generative model's latent space actually contains. The same interface doubles as a debugging aid: because it shows what the model can and cannot produce, you can probe a GAN's limits before committing to it. One caveat: the code dates back to Python 2, Theano, and older GPU toolchains, so expect real setup friction if you want to run it today.