The history of artificial intelligence-generated art, tracing it back to the 1960s when computer programming was first used to generate graphics and images. It wasn't until the recent rapid development of AI technology that generated art began to receive more widespread attention.
The document also highlights the 2015 development of the computer program DeepDream by the Google team. This program transformed images into abstract works of art, causing a stir on the internet and leading the wave of AI-generated art.
The most common AI algorithm is the Diffusion model, used by many platforms that require "prompt" for drawing. The model adds noise to an image until it is unrecognizable and blurry. The AI then reconstructs the original image using step-by-step methods. If the AI can decipher the correct image from the chaos, it shows the ability to create something from nothing.
CLIP, or Contrastive Language-Image Pre-training, connects text and the Diffusion model. It tells the AI what to see and how to correspond text with relevant images. The AI painting process follows these steps: we input prompts and CLIP produces a corresponding representation, A. At the same time, the diffusion model randomly generates an image and CLIP gives a corresponding result, B. By calculating the similarity between A and B, we make A and B infinitely close, eventually achieving A=B. This means that the input description is exactly the same as the computer-generated result.
The use of AI-generated artwork for commercial purposes is currently unrestricted due to the absence of clear copyright laws and regulations. However, creators should be cautious when using any materials that may infringe on the copyright of others, as this may raise concerns of copyright infringement. To avoid legal ambiguity, creators are advised to maintain respect for original works throughout the creative process and seek legal advice if necessary.
DALL-E is an AI program that generates images from text descriptions. It was developed by OpenAI using a 12-billion-parameter version of the GPT-3 Transformer model. While the source code has not been released, demonstrations of text descriptions and their corresponding generated images are available. The GPT model was developed by OpenAI in 2018 and has since been scaled up to GPT-3 with 175 billion parameters, released in 2020.
A 2D female character in a semi-side 45-degree pose, with a dreamy blue background. The character has pink hair and wears a crystal crown on her head. She is making a salute gesture with her hand and wearing a pink JK jacket.