When I first got access to DALL-E, it was not only an amazing life experience but since I got in early (thank you again OpenAI team), I was given access to the tool in a pretty much unlimited capacity. This meant, specifically, I would receive 9 image results at a time, unlimited prompt lookups, unlimited variations, and unlimited edits/inpainting. I am still grateful for the opportunity.
What I immediately found was that after several prompt lookups in the platform, I not only developed an intuition for the model’s capabilities, but I felt my imagination and a space of possibilities growing faster and wider than ever before around me. I was constantly thinking of prompt ideas and had several lists going at the same time, exploring all of them as fast as I could type!
But then OpenAI placed restrictions on use, limiting prompts to 50/day. Crash, it seems that many of his favorite renders
… and I realized they were all the result of so many later stage prompts which I entered into DALL-E. They were outputted very well past the 50 prompt per day limit. At some point, you are so deep into prompt lookups and iterating so fast, you can’t necessarily even trace back how you got to where you are. You’re rapidly exploring new territory, iterating on prompts/ideas quickly, and at the upper limits of your imagination and mental visualization abilities. All of these above images were not what I had initially set out to make when I started entering prompts into DALL-E, but through a longer process of iteration and discovery, I was able to discover them. What I believe is my best and most original work which I also enjoyed making the most.
However, since the quota, I have not been able to reach this important state of mind. At first, I struggled to find an adequate word for this stage, but then I realized it already exists for other creative disciplines - artists, musicians, programmers, writers, all describe the importance of reaching some kind of “flow state” where they can work at their peak creative levels.
Is there an equivalent state for the AI engines themselves? What would the mechanism be? In the case of humans one can imagine a tuning of the nervous system through recruiting more neurons to the task and a shift in neuro-chemical concentration. What would be the equivalent in an AI model?