KYM Review: The Year Of AI Generators

When we were writing our review for the breakthrough meme formats of 2021 last year, we included AI image generators such as Artflow.ai and NeuralBlender on our list. What we didn't know back then is that in 2022, we'd be dedicating an entire review to AI image generators, which impacted meme culture to an unparalleled degree this year.

Powerful image-generative artificial intelligence (such as Stable Diffusion, DALL-E and Midjourney that are trained on the hard work of thousands of real artists and able to create nigh-photorealistic imagery) have turned out to be "Pandora's Box" moment for the world of art — and while artists are making a stand against machines regurgitating their honest work, it is our job as meme researchers to review the way the AIs have influenced the way we memed in 2022.

Also, this review has not been written by an AI (yet).

The Big Three: DALL-E, Stable Diffusion and Midjourney

When DALL-E 2 got on our radar in early May 2022, many of us found the things it was able to do simply mind-blowing. Photorealistic imagery created from textual prompts meant that a part of the glorious AI future we were promised had finally arrived. It wasn't funny yet, but it was impressive.

The hype for DALL-E didn't even begin to subside when the release was followed by two AIs that were just as impressive — Midjourney and Stable Diffusion, both bringing us gifts of memery.

"Stable Diffusion:https://knowyourmeme.com/memes/sites/stable-diffusion was the first of the "Big Three" to find its meme niche — being able to generate images from a visual prompt (another image) meant that Stable Diffusion was able to reshape, let's say, Pepe the Frog, into something that looked very much like Pepe the Frog, but at the same time was a completely different image. Yes, that Among Us parrot that keeps getting posted under every viral Twitter post was made exactly this way.

Later in the year, Midjourney became the talk of the month thanks to its updated image prompting feature, which allowed merging two images into one (much like FaceApp Face Morphing but for any two images).

The Comic Relief: DALL-E Mini and AI-Generated Greentexts

Powerful and divisive as they are, neither DALL-E nor Midjourney nor Stable Diffusion were the biggest AI in memes this year. This title rightfully belongs to DALL-E Mini, which arrived during the narrow window of time when the word of image-generating AIs had already gotten out, but the threat they posed to artists had not yet been fully realized.

During these carefree days of June, it became the entire Internet's favorite past time to come up with the most ridiculous prompts to feed DALL-E Mini (now known as Craiyon) — a generative AI that was not as powerful as DALL-E, but freely available to the general public. Gender reveal 9/11? Done. RGB Gaming Guillotine? No problem. Jesus checking out memes on his phone? Will take but a minute.

But DALL-E Mini wasn't alone in its noble business of entertaining online audiences of all ages. Also in June, Twitter and 4chan discovered that you could use the GPT-3 language model to generate greentexts — the form of sharing personal anecdotes popular on imageboards. GPT-3 captured the spirit of the expectation-subverting format so well that /r9k/ might as well now be filled with actual robots and we wouldn't know any different.

The Writer: GPT-3

Making greentexts was far from the only application inquisitive users found for GPT-3 (or Generative Pre-trained Transformer 3). Able to generate writing from a provided prompt or request, GPT-3 wrote stories, anecdotes, jokes, prose, poetry and haiku, simulated imaginary friends and average Redditors. The New York Times said it could write original prose with mind-boggling fluency. It could probably have described itself far better than we can. Exciting, and also scary.

GPT-3's potential was tamed and fine-tuned into something a bit more user-friendly in Narrative Device, a story-writing bot powered by the model. Generating one-paragraph shorts about your favs? Count us in for that.

The Ones to Talk: ChatGPT, LaMDA and Replika AI

They can write, they can paint, but what else is on the menu? How about chatting with people and maybe passing a Turning Test or two in the near future? In the next-level chatbot department, we have ChatGPT (based, you guessed it, on the GPT-3 model), Google's LaMDA (definitely not sentient) and Replika AI.

ChatGPT, made available just a few weeks back in late November, has already blown some heads with its speaking abilities that are, in all honesty, indistinguishable from a real human. The bot, which can, for example, explain the worst-case time complexity of the bubble sort algorithm with Python code examples in the style of a fast-talking wise guy from a 1940's gangster movie, is nothing short of impressive.

OpenAI's new ChatGPT explains the worst-case time complexity of the bubble sort algorithm, with Python code examples, in the style of a fast-talkin' wise guy from a 1940's gangster movie: pic.twitter.com/MjkQ5OAIlZ

— Riley Goodside (@goodside) December 1, 2022

Then we have LaMDA, a chatbot developed by Google that one engineer infamously claimed was sentient, sparking a viral debate on social media earlier this year. It's definitely not anything close to sentient, but the concept of AI feeling convincingly human is officially no longer "future" but is instead "today."

And then there's Replika AI, which simulates your friend in life, and simulates YOU after you die. Blackmirrored. Blackmirrored again.

The Horny One: Hentai Diffusion

We have to give props to the devs of DALL-E, Midjourney and the rest for trying to install safeguards against the abuse of their creations, but let's face it … people are inevitably going to use AIs to make porn, and there is no way around it.

For example, there's a collective 4chan project named Hentai Diffusion whose goal is to come up with the perfect prompts for Stable Diffusion to spew out NSFW art of large-breasted anime ladies on demand. There are threads upon threads dedicated to the art of coercing AI into drawing well-endowed women, which shows that AI can not only divide but also unite people for a greater goal.

The Anime Kid: Different Dimension Me

Since the dawn of time, people asked themselves the same question, "What if I was a cute anime twink?" Now, thanks to the power of AI (albeit one with some racism issues), we finally have the answer and now are able to experience the wonderful world of anime sans painless assistance from the Truck-Kun.

Meet Different Dimension Me, a Chinese Stable Diffusion-based bot that turns your portrait into your anime portrait in a matter of seconds. Isn't technology wonderful?

---

Looking for more of this year's best viral phenomena and memes? Be sure to check out our other 2022 meme roundups below:

• The Top 10 Memes Of 2022
• KYM Staff Picks Of 2022
• The Slang of 2022
• Breakthrough Meme Formats Of 2022
• The Year Of AI Generators
• Top Gaming Memes Of 2022
• Wildest Stories Of 2022
• Top Streamer Moments Of 2022
• Top TikToks Of 2022
• Top Lewd Memes Of 2022
• Trolls, Rumors and Hoaxes Of 2022
• Top Exploitables Of 2022
• Top Meme Creators Of 2022
• Top Meme Revivals Of 2022