Experiment with chatGPT & DiffusionBee
I am curious about how people will interact with AI.
One pattern I’m interested in is how people will find inspiration.
You come to a machine with no ideas but want some.
How might that interaction work?
For a quick experiment, I used chatGPT as an interface to get prompts to feed into DiffusionBee (note, I say ‘feed’, but i mean a lot of copying & pasting).
1st & 2nd round — science fiction & spooky
I ran into problems early on with the prompt being too long.
TIP: ask for the response to be within / less than a certain word-count
I asked it to tell me some spooky descriptions, and the results I got back were ~kind of spooky, but nothing grabbed hold of me.
One image I liked more than the rest, and it will be used as a personal reference if I’m looking for ‘spooky woods,’ and I like the color palette. The simple fact I found one image that generated from prompts I didn’t think of is still mind-boggling.
This unique image sprang to life from robot words.
Even if there was only one image that I liked overall, there’s still inspiration to grab from the others, whether it’s a texture, color combo, or just an idea that makes me think.
I wanted a more defined style but was still determining what I wanted specifically.
High-level, probably ‘science-fiction’-esque.
I asked for artists that the AI would like to see combined for a science-fiction piece.
Within a few seconds, I got the artist’s names + a brief description that the AI would like to see incorporated.
The prompt was too restrictive to give me anything besides variants of a bunch of trees.
I went back to ask a different question in the third round.
What would the AI like to see?
I’ve been messing around with DiffusionBee enough to get a pretty good idea of when to start tweaking prompts versus going in another direction.
Instead of spooky, a more interesting question would be open-ended and positively skewed.
Describe a painting you would like to see in less than 25 words
In this painting, we see a vibrant, colorful landscape filled with life and beauty.
The sky is a deep, clear blue, and the ground is lush and verdant, with trees and flowers of every imaginable color.
In the distance, a mountain range rises, its peaks shrouded in mist.
The painting is a celebration of nature and the beauty of the world.What’s nice about the response is the phrasing is something I never would have thought of.
Putting in that response, mixed with the artists from the previous question, I got images that really resonated.
That’s the part that I enjoy most about generative art, and I think the biggest reason I get an itch to see ‘what might this prompt produce?’ is that the results bring a smile to my face.
It’s a feeling that I remember as a kid using the computer.
I didn’t exactly know what was going on, but I was able to make stuff work in a way that made me happy.
Now, at 34, I get to type some questions and get responses back that can get transformed into images that I would have no way of spontaneously thinking of.
I’m curious about the near-future and the ability for every person to be able to generate { content } to their specifications.
I don’t think it’s a stretch of the imagination to see a product that allows you to log in, scans all your media consumption habits, and then generates a custom show for you.
If you like a particular series, it might be interesting to see the same episodes but told from a different character’s perspective.
Instead of just spin-offs, you could decide how ‘new’ of a show you want to see.
You could continue with an infinite series or get more involved by including whatever biometrics you have handy.
Want to watch a show that keeps your heart rate in a certain zone?
No problemo.
Want to have a show that is on in the background but doesn’t keep you too engaged, structured so that you know you won’t miss anything’?
I could see people starting to have their own streams/channels of content that they’ve begun to generate.
If you’re bored with your content, you can peek at someone else’s.
I dunno. There’s a lot to think about, but I feel like the progress of everything is happening at such a wild fucking scale that we are all going to wake up in a few months and be living in a different world.
The direction for this post kind of detoured at the end.
I wanted to show how I could ask some questions and get some interesting artwork that has never been seen before, but as I’m typing this, it feels like this is a tiny piece of more disruptive technology.
My interest is in looking at incredible images, but I also like the idea of where AI can help me improve my brainstorming & using it as a spark.
But, it’s also going to be sooooooooooooooooo easy for people to pretend like they did something.
I jumped into Craft to check out their AI Assistant (note — you’ve got 30 requests per month).
After typing ‘What is the future of product design?’, the brainstorm feature replied with a list of 10 ideas.
I typed ‘The role of design tokens,’ and the AI wrote a short paragraph that was mostly filler, but was still something.
Taking one of the 10 ideas from Craft, I asked chatGPT to write a blog post about it, and in a few seconds, it did.
There’s a question of ‘quality,’ but I don’t think it’s going to be relevant for much longer.
The rate at which the models improve will mean there’s only a short gap between people getting dog-shit responses to something people would want to spend money on.
An app that can take in all the content I’ve ever created (both in the cloud & locally), include some photos and video of my environment and, snag enough hours of my conversations to be able to predict how I’d likely respond, and bam — I’ve got an app that can generate unlimited content in ‘my style’ in a way that no one would know (if you’ve ever looked back at work you forgot that you did it, I imagine the blur and wondering ‘did I make this?’ becoming even fuzzier).
In summary:
- explored with chatGPT for word-inspiration
- took word-inspiration into diffusionbee
- viewed results
- iterated
- asked a more open-ended question to chatGPT
- really resonated with the image results
- wanted to share the images and process but got sidetracked by this larger, heavier feeling of the shift that AI is going to bring when it’s accessible to everyone and the larger product ethos of ‘move fast and break shit’ + ‘everyone should monetize whatever they enjoy immediately’ + ‘just release it and see what happens’ will result in some weirdness
I’ve got a link to the images that were generated and hope that you find some inspiration.
Best,
Kelsy
dropbox link to images
I’ve got the generated images saved to dropbox (I list the full url for accessibility reasons).
They are grouped into aispooky and aihappy .
🔗 Other Links
- Craftai (writing application with AI already in preview mode — you get 30 requests on the free plan. Kind of like Notion, worth poking around)
- https://www.craft.do/blog/craft-assistant-future-ai - NotionAI — I’m really liking the approach and am curious about what the cost for the service will be && if demand will have it spin off into its own type of product / service
- https://www.notion.so/product/ai?wr=23b0e9f7ffdff3ee - Verge article on ai’s commercialization
- https://www.theverge.com/2022/12/8/23499728/ai-capability-accessibility-chatgpt-stable-diffusion-commercialization - WashingtonPost article where the author gets a response in their own writing style (I could only imagine how weird it was to write that article)
- https://www.washingtonpost.com/opinions/2022/12/07/openai-chatgpt-writing-tool-journalism-truth
