The future is already here...
One of my favourite William Gibson quotes (but exactly when he said it isn't clear).
The future is already here — it’s just not very evenly distributed yet.
It gets more and more true each day. Take the amazing ability for computers to build images from text prompts:
- January 2021, DALL-E was revealed by OpenAI in a blog post
- People with access? 100's at most? - July 2022, DALL-E 2 entered into a beta phase with invitations sent to 1 million waitlisted individuals.
- People with access? 1 million - July 2022, MidJourney entered open beta, anyone was able to join in and try it out.
- People with access? Early adopters, but limited credits/pay per use - Late August 2022, Stable Diffusion releases code and model weights
- People with access? Most engineers (with python skills + determination) - Early September 2022, single click installers for Stable Diffusion become common place.
- People with access? Anyone with a modern pc and a passing interest
I probably missed a lot with the above, but the point is how fast things are moving from the lab into every technical person's hand. These last few months have been huge, the sheer number of images being created is mind-blowing. And the images are good, in some cases winning art competitions.
The code being on your own computer, no quotas, nothing to sign up to, nothing stored on the cloud - it's freeing, it lets you experiment. The cost per experiment till this point was an opportunity cost, if I try this, I might not be able to try another idea later. It stopped me playing.
But now, with it installed locally, the only cost is time (and a bit of fan noise/electricity). I've already generated over 100 in the last 24 hours, just casually messing about! By comparison, I think I made under 10 across a few months with DALL-E and MidJourney.
Do you want in? Getting your own AI artist is easy!
- Mac M1/M2 - https://www.charl-e.com/
- Windows / Linux - https://github.com/cmdr2/stable-diffusion-ui
cmdr2Or find any number of other one click solutions (new ones springing up all the time):
Writer's block?
Writing prompts for AI image generation is a skill all of its own, there are keywords you can add to get the results you might want. But even just having the inspiration for what to create can be a challenge (unless you have small children on hand, they are FULL of ideas).
Luckily, someone already thought of this, https://lexica.art/ is a search engine for over 10M+ Stable Diffusion images and critically, as well as the image, they stored the prompts which generated them.
A database of 10M+ images is a huge number an impressive resource on its own - but now consider how long Stable Diffusion has been available to the public - just a few weeks!
This HUGE database of content, which took only few weeks to generate, makes me wonder about business based on licencing image based content; but I have no doubt video and audio will follow.
It is not just content licencing business which are at risk or affected by this new, world of fast and easy AI art - "real" artists are up in arms that their digital forms, networks and meeting places are being buried under a tsunami of digital art and are trying to ban it.
Andy Baio
What's next?
Creatives are only just starting to see what they might be able to do with this tech (which is itself still in its early stages). And, for me, the amazing part of all this is the means of production is getting into everyone hands, thanks to open source!
With the pre-trained AI tools, with natural language interfaces, and easy to install deployment - the cost of experimentation and failure is basically zero, and so are the barriers to entry. I don't see this trend going anywhere - and the fight against it, I suspect, is a loosing battle, as it was with digital photography.
Just thinking about what the digital camera did for photography, or YouTube for video distribution makes me think that AI based creations tools are set to disrupt and decentralize even more types of content, only much, much faster.
My kids will have a blast playing with this.
