CVPR 2020: Synthesizing High Resolution Images with GANs

CVPR 2020 dcv20
Presenters: Tech Demo Team, NVIDIA
Developed by NVIDIA Researchers, StyleGAN2 yields state-of-the-art results in data-driven unconditional generative image modeling.

Watch this session
Join in the conversation below.

Supposing this proposed method will eventually work well on photographic inputs, artists going forward will eventually no longer make a profit from generating new subjects from their initial style.

Now, I understand the intention here is to speed up time-consuming processes in order to optimize profit margins at the enterprise level. I help automate processes for companies as well. That is the essential value we all provide as employees.

I’m curious if the intention of the creators of this project is to annihilate the essence of the phrase “artistic style” and reduce stylistic families to a meditation as representational art has become in the place of photography for all of time?

And if this wasn’t the intention because the thought was not raised, what can we all do now for all the artists who will be affected?

It’s always a good idea to have constructive debate on how new technologies might be used, for art or any other application. Other research into new capabilities for creative applications has drawn widespread praise (Over Half-Million Images Created with GauGAN AI Art Tool | NVIDIA Blog) from both the research and artistic communities, and artists have commented positively that new tools have allowed them to think differently and create their visions in new ways. We are certainly open to input from our developer community, so if others have questions or concerns you can contact me directly at

1 Like

The GAN you are referring to is a concept-generator tool as opposed to the tool this post is presenting which will directly replace existing job functions. Moreover, Creative Directors (who were the ones who have commented positively in your article) are the project manager equivalents to development teams who are not themselves the artists doing the work, and can hire and fire artists to do work as they please.

I do not mean at all to oppose but to raise the question that when a tool is released, what kind of instruction manual will be released with it? What resources will be created alongside your development of these tools for those who will eventually be impacted to let them know to pivot their job functions?

A lot of people complain about the rich getting richer and the poor getting poorer. If we are to continue to complain, I hope that when we create tools we are not waving it around legitimizing the greatness of it only because of the technical achievements it provides, but we are also thinking about the long term impacts.

This concludes all that I would like to express on the topic.

1 Like

Hello. Is it possible to play around with the paintings GAN tool demonstrated in the video ? I’m not referring to GauGAN, but this specific one. Is it very curious, but I can’t find any additional information on it except this video.


Last year I was playing with GauGAN and I had an idea to generate animated input maps to create an AI visualization in animated format. I made a YouTube video detailing my process and in the last week have had almost 1 million views! The output was fascinating but I had to click the load, process and save buttons 700 times to create a 30 second video. I’m excited to do more but could anyone create a method to batch process PNGs into the web client or a desktop client?

Check out my video at around the 6 minute mark to really be amazed!