Hey everyone, I recently gained access to DALL-E and decided to see what it could create if I fed it a description of our current logo. I think the results look absolutely fantastic, what are your thoughts? For reference, this is our current logo: https://twitter.com/Team1716/photo
The description I used was “Side profile of a fierce robotic red bird head”
The trick I’ve found is to be hyper specific. I’d also like to note that my images were created by DALL-E by OpenAI (currently in beta), not DALL-E Mini (which was built by someone else). They use the same fundamental techniques, but OpenAI has the advantage of a gargantuan amount of compute to train their AI’s on. I’m gonna look into using this tech for creating ideas for protoypes for next year’s game. Would be kind of neat to say that “Our robot was designed by an AI”.
With various inputs into DALL-E Mini:
Why not the whole robot all at once?
Because we don’t know what the purpose of those robots are (insert butter robot meme here). Though I gotta say those images turned out a lot better than I would have expected from that kind of prompt. IME it’s a bit of a balancing act as to how specific you can be (technical designs) vs general concepts (robot) to get good images.
It’s kind of hard to precisely explain what I’m trying to do, but in the same way you can describe to DALL-E a general image to create and it creates it, I’m trying to describe to DALL-E a general engineering problem and it creates an image of a mechanism that can solve that engineering solution. Now that I think on it, DALL-E might be the wrong kind of tool for this problem, but aspects of the architecture might be useful in creating a tool that can solve this.
I tried this with craiyon.com and the results were interesting:
The faces…
Hak wak is like k-rad with more AI.
that’s a big Silverhawks vibe. I dig it.
I gave it a shot…
Process I used: As others suggested, I used a quite detailed set of words to start. Then I picked an image and used the beta image editor, erasing the head off the body because I didn’t like it, but I liked the rest. And then I added a generation frame (extends image footprint in one direction) and entered text for the spartan head I wanted to add, which resulted in what you see except the colors were green and orange. So then I downloaded that image and made some more edits including slapping our team’s logo on the back of the laptop.
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.
