June 3rd 2022
DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language.
Click on the text descriptions above to select the natural language sentences that were used to generate the “mind-blowing” images.
DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language, combining concepts, attributes, and styles.
It was announced to the general public in January 2022, and is the successor to the first version, released one year earlier. Developed by the company OpenAI, which is a San Francisco-based research laboratory founded by Elon Musk and others in 2015.
According to their website, OpenAI have limited the ability for DALL·E 2 to generate violent, hate, or adult images, by removing the most explicit content from the training data.
The creative potential of this technology is fascinating, but I can’t help being concerned about its ultimate misuse, such as deep-fake videos spreading propaganda. Hopefully a mechanism to determine provenance will develop alongside.
How does it work?
DALL·E 2 uses the language prediction model GPT-3. It has been developed further with 12 billion parameters that “swaps text for pixels”, trained on text-image pairs from the Internet.
It uses zero-shot learning to generate output from a description and cue without further training.
DALL·E 2 can do so much more than the examples demonstrated above.
- It can make realistic edits to existing images from a natural language caption.
- It can add and remove elements while taking shadows, reflections, and textures into account.
- It can take an image and create different variations of it inspired by the original.
For many more examples jump over to their website here.
GPT-3 is the third-generation language prediction model created by OpenAI, a San Francisco-based artificial intelligence research laboratory founded by Elon Musk and others in 2015.
GPT-3’s full version has a capacity of 175 billion machine learning parameters and was introduced in May 2020. It is part of a trend in natural language processing (NLP) systems of pre-trained language representations.
The model has been trained on various datasets such as Common Crawl, WebText2, Books1, Books2 and Wikipedia, amounting to hundreds of billions of words and is capable of coding in CSS, JSX, Python, among others.