Thread by @minxdragon

This work is set to fetch the highest price point for a machine learning based work so far.

I would like to know more about it, I’ll do some digging.
But first, let me tell you my feelings on this work... christies.com/features/A-col…

Here is the entire painting from @ChristiesInc. It has an unfinished quality to it, blank canvas, interesting use of white space. But, for me, the main success is in the texture. This has a painterly feel that represents the era but with a modernity to the composition.

Most GAN portrait artists complain about the eyes in their generative works. They are either dark or artifacted, it’s hard to produce compelling eyes.

These are dark and brooding. Like holes into his soul.

This work misses features, the face is indistinct. We see that often with GAN generated works, unless you provide a semantic map or a base to build on you are going to be missing bits and blurring.
But for all it’s indistinctness it has a faceness to it. You know it’s a portrait.

I think sometimes we want to make something exactly like the source material. We want it to be exactly like an 18th c portrait. But to do this misses the opportunity to explore new modalities of abstraction and interesting ways to look at things

Success in a work isn’t about emulating it perfectly, although that is a skill in itself. The success in this piece is in the materiality, in the essence of the portrait and in the slight abstraction in a way a human painter would not abstract a work.

I think that @obv_ious have created something unique in this work. Not just this one painting, but the collection and the family tree. Some are more realized than others but this work is an example of a later style, hastily worked wet on wet with missing parts, a sketch.

The phrase “poorly trained GAN” was bandied about but to me this reflects a similar sentiment to “abstraction is for people who can’t draw”

And this is where the artist comes in.

The curation of the works is very much part of the art here, curating the source images, curating the outputs, putting them together in a cohesive narrative. You have to know the subject in order to abstract or simplify it.

A GAN and the artist working with it understands the subject. They have to.

After curating thousands of images to put in your dataset I guarantee you see them in your sleep. Does the GAN? Well that’s a question for another day.

Am I a little bitter that it isn’t my work? Maybe. I could use money like that, and kudos like that. But it all draws attention to our work. It all legitimizes our practices. We need to build each other up not tear each other down.

There is room for all of us.

My current work would possibly also be considered the results of a poorly trained GAN. And they are to some extent. But what I am looking for isn’t a perfect creation of a new sculpture although maybe one day I will. The works that stood out for me had bodiness to them

The works had a sculptural aspect and a human form that I could see and work into and find the sculpture in the rough output.

And that together we would create something extraordinary.

They are rough. But I wanted to explore that roughness. Just as a marble sculptor sees the form inside the marble so to do I see the form inside my GANs output.

I will be examining this further and realism will be one of the areas I take it to, but I wanted this abstraction. I embrace it.

Most of creating art is in the intent. @obv_ious had a clear intent with these works and it shows. And that is why they are successful.

@threadreaderapp unroll please