I wanted to share my process for creating images and text that go together for projects such as Choose Your AI Adventure. First we’ll look at Claude-to-Midjourney, then we’ll look at Midjourney-to-Claude.
The two entries in the current round of Choose Your AI Adventure were made through these processes. Please go vote now and give an entry a shot for the next round which will begin tomorrow!
The second place story for Choose Your AI Adventure was “Shades of Silicon.” We’ll use this intro and pretend we’re going to make entries to continue it. Here’s the introduction:
Claude-to-Midjourney
This time we’ll start with text. I gave Claude a brief explanation, the story so far, and instructions as to what I was looking for. Here’s the prompt:
(Note: With “The Broken Lands” entries, I gave it the whole text as it’s been built so far. Claude can handle alot of data, so don’t be afraid! You can even just give it a txt file.)
Let’s take a look at what it came up with!
Not bad! I’ve always been impressed by Claude’s responses when it comes to stories, though I will say sometimes it seems to ignore the word-count requirement. Personally I like the second one.
Once I have something I like, I look for the image I want to create. In this case, the drone patrol really strikes me.
Time to head into Midjourney.
It took me a few tries to come up with something I liked. Here I really liked the fourth option.
Now, if I want, I’d go back to the story and edit it. I might want to focus more on the drones and maybe stop the scene right after the escape from them. Either way, I have something I can easily mold into an entry.
Midjourney-to-Claude
Now to start from an image. Claude was heading in a good direction with all the security, so I decided to go with that — I want an image of security guards in front of the tower. I also want to see them through some kind of overlay so you really get that cyberpunk feel.
After a few tries I ended up with these…
Now that I got my image, I head over to Claude. I like starting a new instance with these, that way it can get confused by the entries it already created. Everything the same, except for the instruction bit — this time I give it details from the image that I want to see in the text:
”Please write an 150-200 word entry for what happens next. It should feature security guards in front of the tower, as seen through a cybernetic overlay.”
Here’s what I got back:
Again, very impressive! I have a chose here either to go back to Midjourney and try to adjust the details to match the text, or edit the text. I’d probably just edit the text, using it as inspiration.
Conclusion
That’s it! If you want more control over the writing, you can ask Claude to help you brainstorm ideas instead of writing the entries itself. Similarly to brainstorm on Midjourney you might enter a vague prompt like “cyberpunk, streets, danger --c 100”, see what it comes up with, and narrow down the images you like.
Was this explanation at all helpful to you? What has your workflow been for projects such as this?
Comments