[octothorp]

The last month has been absolutely insane for the state of AI-generated imagery... throughout this year Dalle 2 has been gradually opening up their service to an increasingly large number of users, Midjourney released their v3 last month and quickly maxed out the number of users that their discord-based service can support, and this week Stable Diffusion released their open-source model for anyone to download and run on their computer (if your computer can handle it). What seemed like a novelty a year ago seems like the inevitable future for a lot of creative industries... if not outright replacing people, at least vastly changing workflows.

(If you're not familiar with these technologies, they all operate in a similar way: you write a sequence of words, a prompt, which the algorithm than used to create an image for you. The algorithm has been trained on anywhere from hundreds of millions to billions of images, which it uses to figure out what you want. Results range from photorealistic to artistic to just plain weird and incoherent depenting on the algorithm and what you ask it for.)

Each of these services has their own strengths... Dalle 2 is fantastic at creating realistic images of things that don't exist. Midjourney is the more artistic, being absolutely fantastic at capturing the styles of different artists, but struggles at coherence. Stable Diffusion sits somewhere between the two, but has the advantage of being locally runnable and open-source.

There are so many interesting societal and ethical issues involved in this. These services haven't paid for the rights to use the images that they use for their dataset, and in the case of SD in particular, sometimes it'll return an image that is more like a photo-collage of existing photos... sometimes even showing watermarks from clipart sites on the final results. Midjourney does a fantastic job imitating the styles of other artists... but is it ethical for someone to type in another artist's name and get something that imitates that style? Does it matter whether the finished result is used commercially, or whether the referenced artist is still living or working?

Then there's the whole 'objectionable content' side of it. Both D2 and MJ attempt to control their created content to limit the amount of nudity, gore, celebrity fakes, etc. being generated. SD has no such practical limitations (just some vague and unenforceable terms of service), and when MJ did a beta-test that used SD's dataset earlier this week, they shut it down after a day because the extreme objectionable content being generated was more than their mods and development team could manage. Reddit has already banned a large number of SD NSFW communities because of the content (particularly the faked celebrity photos). It seems likely that there within years there will almost certainly be web-based services aimed specifically at the NSFW audience, because that audience is huge and willing to pay.

In artist community discussions, the discussions range from dismissing it as completely not a threat to their industry; to freaking about about becoming obsolete quickly. The middle-path approach seems to acknowledge that these are tools that will inevitably change the industry and how artists work, but won't fully remove artists from the process. I think the real change will come once entertainment studios from Blizzard to Disney have their own in-house models tailored to their needs... something that's already almost certainly in development. And it won't be just 2D static images: Midjourney already has a 3D version in the pipeline, and while animation is beyond what these public models can currently do, big studios with the money to spend on the hardware could absolutely do that.

My own experience has been primarily with Midjourney, although I also played around with SD's model earlier this week when MJ had a brief beta-test of SD's model, and these are tools that I'll likely involve in my own creative workflow going forward. I've got my Dalle2 invite but haven't used it yet, although later this year I'll definitely explore some of its tools and see how they can be used in tandem with MJ. I do have societal concerns, like if in the future will there be fewer kids learning to draw, if you can get a computer to render whatever you imagine. Does it even matter if fewer kids are learning to draw?

I'm curious to know if anyone else is exploring these tools, or thinking about what the implications are.