Synthetic Images
The pope's new coat from Balenciaga makes him look more like P. Diddy than P. Franny, no?
Oh wait. That's not a photograph of the pope. It's a deep fake.
It's a composite image made (1) ostensibly by a construction worker in Chicago, Pedro Xavier (2) using off-the-shelf AI image generation software, Midjourney (3) fusing together photographs that other people took and posted in the dataset on the internet.
This image is worth pausing on, to think about for a moment.
Despite all the magic talk about Chat GPT becoming sentient and Midjourney winning a photography competition, the singularity is not here yet.
Nor even the runaway explosion that is supposed to precede it.
But things seem to be accelerating, and specific to synthetic image generation we seem to be at some type of inflection point based on the amount of interest (and pictures) being generated as-of late.
I don’t have much direct experience beyond the Adobe Firefly beta, and so for now will just share a few takes and pieces that I think shape the context around the public conversation.
Can We No Longer Believe Anything We See? An NYT explainer piece that ranges from the Bennett Miller show at Gagosian, to the launch of Adobe Firefly, to the Getty Lawsuit against Stability AI, the developer of Stable Diffusion. And the implications of all of this for the moment and the future.
https://www.nytimes.com/2023/04/08/business/media/ai-generated-images.html
A Look At How Text-To-Image Algorithms Work. A five-minute explainer video featuring Google AI research scientist Laurence Moroney describing the main methods of TTI including Diffusion (noise) and Autoregression (tokens).
https://www.youtube.com/watch?v=9YrYDqhJdPw
The Best Synthetic Media Apps (March 2023). A comparative roundup of the best synthetic media apps - video, image, text, voice - from technology consultancy G2, plotted onto their familiar quadrant. Includes all of the household names including ChatGPT, DALL-E 2, Midjourney, Stable Diffusion - and many others.
https://www.g2.com/categories/synthetic-media
Twitter Synthetic and Manipulated Media Policy. This article from Twitter's Help Center explains the platform's policy on synthetic and misleading content and potential penalties against users who create and share it, including content labeling, content removal, and account lockout.
https://help.twitter.com/en/rules-and-policies/manipulated-media
The Content Authenticity Initiative. The CAI is a community of media and tech companies, NGOs, academics and others working to promote adoption of open industry standards for content authenticity and provenance.
https://contentauthenticity.org
Why Pope Francis Is The Star Of AI Generated Photos. A look at why pictures of Pope Francis in a puffer coat (above), at Burning Man, on a motorcycle, fascinate us. The formula according to a Yale Divinity School professor interviewed in the piece is: familiarity x incongruity = defied expectations.
https://www.nytimes.com/2023/04/08/technology/ai-photos-pope-francis.html
