Feedback-Loop-Art
A modern take on Vermeer

Feedback-Loop-Art

Hasn't art always been about delving into the unknown? Venturing into something for the very first time in pursuit of deeper meaning or simply for the sheer joy of creation?

When I first encountered images produced by the new Midjourney Version, I thought, "That's it, I'm setting my camera aside. Why exert the effort to capture photos when you can summon them into existence with a simple prompt?"

Navigating through this initial disillusionment took some time, but I've since emerged, reinvigorated and eager to explore uncharted territories. Or, at the very least, re-envision ideas that have been around for ages.

Thanks to GenAI, I can now experiment with novel concepts in mere hours rather than months. Just this evening, I dabbled with the idea of Feedback-Loop-Art. I'm uncertain if it gained traction yet. When I outlined my vision to GPT4, this was one of its suggestions. So, how does this process unfold?

DESCRIBE, VISUALIZE, REPEAT

Basically it's a two step approach involving multimodal generative AI like GPT4.

DESCRIBE: Upload a picture and ask the AI to describe it as best as possible.

VISUALIZE: Generate that picture with the description by the AI itself and repeat.

I began with the renowned painting by Vermeer, "The Girl with a Pearl Earring." After numerous iterations, here are the outcomes.

The Original, the first iteration, the last iteration

Essentially, by engaging in a game of "telephone," subtle distortions are introduced, leading to entirely new results. These transformations occur progressively, as evident throughout the initial session:

14 Iterations of the Feedback-Loop Game

To establish it as a novel art form, I've refined the prompts that originally provided the descriptions.

DESCRIBE: Upload a picture and ask the AI to describe it as best as possible and add some artistic freedom*

VISUALIZE: Generate that picture with the description by the AI itself and repeat.

*the actual prompt: Please describe the following picture as detailed as possible so that it can be recreated with a text 2 image generator. You are meant to improve the picture to make it much more captivating and modern. You are not asking for further input. Output format should be a few sentences.

And as you can see the results are vastly different:

The Original, the first iteration, the last iteration

As evident, there's a significant evolution throughout, with each image standing out as a unique piece of art.

That concludes my showcase for now. Which image resonates most with you? Have you encountered similar concepts before? I'd love to hear your thoughts on it.


Here are some high-resolution generated artworks for your viewing pleasure:


要查看或添加评论,请登录

社区洞察

其他会员也浏览了