Many models on realistic face generation are being experimented with, where the data sets have been culled from the Imagenet and similar public databases. While there is a remarkable effort to fit the realistic faces onto the input prompts, most of the models showed skewed datasets, with biases like the proportion of asian, global south etc being much much lesser than the global distribution etc. Also there seems to be a deep desire for mastering realism so models can produce images that might be undifferentiated from reality.

This overfitting of limited parameters to achieve perfection belies the dark truth of human endeavor that encourages achievement without addressing inclusivity, priviledge and the core issues first. And the fact that imperfection might in itself be perfect is also a completely missing from this narrative.

This series explores this publicy available image generation model to generate faces through drawings and by experimenting with changing the parameters of the model to generate a set a faces based on the inputs. Rather than achieving perfection, the attempt is to push the imagination curve higher. The publicly available model used was deepfacedrawing.geometrylearning.com