Flipboard Friday: AI Banksy Art Contest¶
Source: https://www.wte.net/Blog/April-2023/Flipboard-Friday-AI-Banksy-Art-Contest
Date: April 2023
Author: Martin W Smith
Introduction & Inspiration¶
The author references a previous popular post about Banksy's guerrilla marketing techniques and decides to explore using artificial intelligence to visualize the mysterious street artist. He shares a personal narrative about receiving a cancer diagnosis 17 years prior and subsequently creating a bucket list. Among his goals were bicycling across America (accomplished in 2010) and meeting renowned artists. He recounts encounters with Robert Rauschenberg, Keith Haring, and Peter Max—experiences he found deeply inspirational.
AI's Description of Banksy¶
Since direct contact with Banksy remains impossible, the author consulted ChatGPT. The AI language model acknowledged it cannot create images but noted that "Banksy's true identity has been kept a closely guarded secret" with only speculation available. It suggested the artist might be "a white male in his 40s or 50s, with a casual, unassuming appearance," possibly using disguises or working with a team.
AI-Generated Images¶
Using ChatGPT's description as a prompt for DALL-E-2, the author created images depicting this speculative version of Banksy in a hoodie using Shepard Fairey artistic style.
AI Banksy Art Competition¶
Overview: The author invites participants to use AI art generators (Midjourney, Bing Image Creator, Adobe Firefly) to create their interpretations of what Banksy might look like.
Entry Requirements: - Send inline images or secure links - Contest runs April 20 through May 4 - Email submissions to martin@wte.net or mobriff@gmail.com - Non-AI entries (drawings, Photoshop work) also accepted
Prize: Winner receives a sheet of Banksy's Ukraine stamps featuring a girl flipping Putin—imagery the artist created on Ukrainian streets during the conflict.
Judging: Approach depends on entry volume. With minimal entries, the author will select a winner and explain reasoning. High volume might warrant public voting.
Understanding GANs¶
The author explains Generative Adversarial Networks (GANs), describing them as machine learning systems invented by Ian Goodfellow. These consist of two competing neural networks: a generator producing realistic samples from random data, and a discriminator determining whether samples are authentic or fabricated. Through adversarial training, the generator improves at creating convincing outputs while the discriminator enhances its detection abilities. When equilibrium occurs—when generated samples become indistinguishable from real data—training completes. GANs enable realistic image generation, resolution enhancement, style conversion, music composition, and synthetic data creation.