AI generated 150 years worth of photographs in less than a year

from 150 years to 12 months: how ai created 150+ billion images

AI-generated artworks (and deep fakes) have taken the world by storm. From imagining floatable structures and surreal furniture to recreating ancestral portraits, text-to-image algorithms — e.g., Midjourney, DALL·E 2, and Stable Diffusion — have become a new favorite tool for artists. Combining the precision of entering prompts with the creative human vision led to astounding visuals being shared across the Internet. Stock image search engine Everypixel released a detailed report that quantifies that growing phenomenon with a rough estimate: In only 12 months, text-to-image algorithms generated over 150 billion images. ‘Since the launch of DALL·E 2, people are creating an average of 34 million images daily,’ writes Everypixel. According to the report, it took us humans 150 years to amass 10% (15 billion) of that mark, demonstrating the incredible — and equally alarming — power of machine learning.

The Mutated School Bus series by Shail Patel using Midjourney | image © the author (read more here)

full header image © Momit Sam York via Instagram

everypixel quantifies the impact of text-to-image softwares

Everypixel’s report combines statistics and facts related to four different text-to-image softwares: Midjourney, DALL·E 2, Stable Diffusion, and Adobe Firefly. For starters, OpenAI introduced DALL·E 2 in April 2022 initially as an invitation-only platform. Access expanded over several months until September 2022, when it became open to all users without restrictions. By then, two million images were being generated daily. While the exact timeframe and method for calculating this figure remain unclear, ‘we assume that this is an average, which means that approximately 916 million images have been generated on a single platform in 15 months,’ shares Everypixel.

Gaudí-inspired vacuum cleaner by Marcus Byrne using Midjourney | image © the author (read more here)

Midjourney comes next in line. Launched in July 2022, Midjourney’s usage on its exclusive platform, Discord, involves approximately 20 to 40 image creation tasks per second. It boasts 15 million registered users and 1.5 to 2.5 million active members at any given time. Using an average of 30 image tasks per second, the daily count could reach up to 2.5 million. Since its launch, roughly 964 million images have been created with Midjourney.

Meanwhile, in August 2022, Stability AI unveiled its text-to-image model known as Stable Diffusion. Official testing sites for this program include Dreamstudio and Stability AI’s space on Hugging Face, with over 10 million users across all channels. ‘If we extrapolate the Midjourney’s numbers and trends that we have at hand, it turns out that through the official Stable Diffusion channels, users generate 2 million images on a daily basis, and in more than a year since the release, this number has reached 690 million images,‘ the report states.

NEWFACET fuses AI & fahshion using Stable Diffusion | image courtesy the author (read more here)

The challenge, though, lies in Stable Diffusion’s open-source nature, meaning content creation extends beyond official platforms. Numerous platforms, apps, and services built atop Stable Diffusion’s technology further complicate estimation as they cater to a substantial and unquantifiable audience, producing content that’s hard to gauge with constant growth. Everypixel examined repositories like GitHub, HuggingFace, and Civitai to better understand that challenge. These platforms host thousands of Stable Diffusion-based models by users. An initial estimate came up when applying Midjourney’s patterns to Stable Diffusion models on these platforms. Later, data from the Civitai team refined the estimates, revealing over 11 billion images generated from these repositories. When including other popular models (like Runway) and Stability AI’s official channels, the Stable Diffusion image count rises to 12.590 billion, comprising 80% of all AI-generated images using text-to-image techniques.

Gustav Klimt’s ‘The Kiss’ & its AI variations using DALL·E 2 | image courtesy OpenAI (read more here)

Joining Midjourney, DALL-E, and Stable Diffusion is Adobe Firefly, a text-to-image Photoshop feature that produced over 100 million assets within six weeks of launch (March 2023) and 1 billion visuals after only three months. Integration with Adobe Photoshop in May 2023 further propelled image creation thanks to the software’s popularity among global creatives. ‘In total, more than 15 billion AI-created images have been generated using Stable Diffusion, Adobe Firefly, Midjourney, and DALLE-2. That’s more than Shutterstock’s entire library of photos, vectors, and illustrations, and one-third of the number of images ever uploaded to Instagram,’ concludes Everypixel.

concept Apple Car by John Mauriello using DALL·E 2 | image © the author (read more here)

turning to the dark side with AI FAKES

Moving away from the sunny side of AI-generated images, you’ve probably seen the viral images of the Pope in a puffy coat or the arrest of Donald Trump. You may have even seen images of the Blue Plague in Russia and wondered if you were blocking something important from your memory. If you haven’t figured it out yet: They’re fake – AI-generated, to be exact. And while it’s not the first time AI images have gone viral, apparently nothing has ever spread so quickly and widely, and never have such images left so many people questioning whether they’re real or not. In only a few months, generative artificial intelligence has progressed from the photorealistic images of DALL·E 2 and Midjourney to even more new tools that create realistic videos and entire movies using only words. All indications are that artificial intelligence isn’t going to slow down anytime soon, and while it’s admittedly fun to experiment with it, there’s a truth that it poses a certain threat – especially in the context of misinformation.

Read more on designboom here.

fake visual showing the arrest of Donald Trump | image by EliotHiggins via Twitter