4 ms·

I have been experimenting with this system for months and have concluded that it is a very powerful paint brush. Powerful like steering an elephant. Not like be

I have been experimenting with this system for months and have concluded that it is a very powerful paint brush. Powerful like steering an elephant. Not like being Superman. The system is garbage in garbage out. So, feeding it random noise inputs will give noise results. The human factor is “prompt engineering” the inputs to try to coax the elephant in the desired direction. And, in evaluating the outputs. Sometimes it feels like asking a child to draw something for you. I spent some time trying to get “storm clouds made of lava” but the child insisted “No, dad! Lava stays on the ground. Like THIS!” https://www.reddit.com/r/bigsleep/comments/o033x1/playing_with_lightning_today/ https://www.reddit.com/r/bigsleep/comments/o033x1/playing_wi... Similarly, while trying to make a “tempest in a teapot”, instead of a storm, the child was a fan of the My Little Pony character Tempest Shadow. https://www.reddit.com/r/bigsleep/comments/o04sin/trying_to_make_a_tempest_in_a_teapot_but_the_ai/ https://www.reddit.com/r/bigsleep/comments/o04sin/trying_to_...

marcus_holmes 5y ago
That's awesome :) Thanks for the info. I love the images. And the metaphor of it being a powerful tool for creative expression rather than an actual creative expression itself. Do you think this will change and get better? Would it help if the AI was capable of interactively refining the inputs (I guess by asking questions instead of making assumptions)?
- corysama 5y ago
  I think that the tool will get higher quality results with easier to control inputs. Some of that is already happening rapidly. In theory, given enough machine power, I could see a branching, interactive interface. Like if there are multiple major hills in the latent space ("Tempest-> a storm cloud" vs. "Tempest -> a cartoon pony") the system could identify them all and generate a set of images with different biases towards each hill. Then the user can pick which direction to climb. The current system on high-end hardware takes about a minute before results for one image even begin to be recognizable. And, about 3 minutes before you can be somewhat confident how it's generally going to turn out.

marcus_holmes 5y ago

That's awesome :) Thanks for the info. I love the images. And the metaphor of it being a powerful tool for creative expression rather than an actual creative expression itself. Do you think this will change and get better? Would it help if the AI was capable of interactively refining the inputs (I guess by asking questions instead of making assumptions)?

corysama 5y ago
I think that the tool will get higher quality results with easier to control inputs. Some of that is already happening rapidly. In theory, given enough machine power, I could see a branching, interactive interface. Like if there are multiple major hills in the latent space ("Tempest-> a storm cloud" vs. "Tempest -> a cartoon pony") the system could identify them all and generate a set of images with different biases towards each hill. Then the user can pick which direction to climb. The current system on high-end hardware takes about a minute before results for one image even begin to be recognizable. And, about 3 minutes before you can be somewhat confident how it's generally going to turn out.