Can Stable Diffusion Disrupt Dalle2? 🖼️
Text-to-image deepfake era nearly here.
Hey Guys,
This is Data Science Learning Center Premium,
Since Stable Diffusion is open-source, it could potentially spawn A.I. that gets around rules of the internet. I enjoyed some of the debate recently on Hacker News here. Midjourney and Stable Diffusion could end up rivaling Dalle2 and it’s a bit embarrassing for the commercial ambitions of OpenAI, who are greatly sponsored by Microsoft.
See Stable Diffusion on GitHub: https://github.com/CompVis/stable-diffusion
Midjourney is a research lab and the name of the lab's artificial intelligence program that creates images from textual descriptions, similar to OpenAI's DALL-E.
I find the mission statements of both Stability.AI and Midjourney.AI somewhat comical. Midjourney is an independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species. Guys, don’t make me laugh!
As for Stable Diffusion, landscape images could likely fool most people!
When users sign up to DreamStudio they will be given 200 credits but after that, £1 ($1.18) will buy 100 generations. £100 ($117.76) will buy 10,000 generations.
I mean as long as Super Mario looks a bit like we imagine right? DALL-E 2 wins this contest for sure:
“Of course, they are not real photos and they only resemble real places (for now, as the input data becomes larger and larger) but for most of the people that consume the images, it is only about the beauty of those images.”
We are very near a deepfake era that will be very difficult for the Web2.0 to handle. But let’s take a look at some of Stability.AI’s landscapes.
A nice glow!
Pretty realistic!
No way! Yep.
Really majestic! Image credits: All above images by Aurel Manea. (PetaPixel)
Some writers are saying that Stable Diffusion embodies the best features of the AI art world (Algorithmic Bridge): that, “it’s arguably the best existing AI art model and open source.” I have no idea really, but it does seem pretty good!
I think however comparing Stable Diffusion, Midjourney and Dalle2 makes some sense at this point.
Fabian Stelzer recently made an image comparison test between three artificially intelligent (AI) text-to-image generators: DALL-E, Midjourney, and Stable Diffusion.
Stelzer published his findings in a Twitter thread where he explained the process that involved giving exactly the same prompt to each machine and using a 1:1 aspect ratio.
The results are seriously amusing:
Midjourney: Painting sketch
DALL-E 2: More realism
Stable Diffusion: Fantasy
If I had to describe them that’s how I would do it in a nutshell.
When I see Stable Diffusion’s body of work, I think of gaming worlds.
Prompting aside, it’s clear that DALL-E 2 while first out of the gate, is not necessarily the best.
We are entering essentially “deepfake for all” territory here.
NovelAI has been experimenting with Stable Diffusion to produce art that can accompany the AI-generated stories created by users on its platform.
Can you see what will happen here? This will proliferate into new apps, new startups, new use cases for AI-generated Art both for good and for potential misuse.
“Stable Diffusion will allow both researchers and soon the public to run this under a range of conditions, democratizing image generation,” Stability AI CEO and founder Emad Mostaque wrote in a blog post.
This “democratization of AI” line is getting really old very fast. Hugging Face really helped Stability.AI become what it is this month. Commercially it could hurt OpenAI.
It’s a brave new world for deepfakes at scale to corrupt the platforms, like a virus.
How will we ever know what is real again? Stable Diffusion is very much new territory. Other AI art-generating systems, like OpenAI’s DALL-E 2, have implemented strict filters for pornographic material. TechCrunch sounded upset.
So did Emad Mostaque learn from OpenAI’s mistakes? And will these new startups like Stability and Midjourney test the so-called dominance of OpenAI, that doesn’t feel quite as special any longer? Stable Diffusion 🎨 is here to stay.
Join 33 other paying subscribers for additional content and locked archive posts.