GBKS on Nostr: Thanks. Yes, consistency is super hard. It takes a ton of iterations to find things ...

Thanks. Yes, consistency is super hard. It takes a ton of iterations to find things that match. For these, I used the Niji style in Midjourney, which is trained on anime. Then I put a few specific style hints at the end of all prompts, in this case the names of a few anime artists I picked out that work well. Another part is being very specific with image compositions. For example, I try to have only one close-up of a specific character, because I know I can never get another one that is close enough. So for other images, I try to think of compositions where the overall environment or scene take more focus than the facial details. Overall though, you have to draw the line somewhere and live with a certain amount of inconsistency. And my hope is that the tools get better and future iterations can iron this out. If you want to peek in, this Figma file has all generated images for this project (also ones not used) along with their prompts. [https://www.figma.com/file/0Fg2IiEbLI5o8juawKqkJM/Saving-Satoshi-images?type=design&node-id=407%3A1645&mode=design&t=vuHD8KdxXnXQER8K-1](https://www.figma.com/file/0Fg2IiEbLI5o8juawKqkJM/Saving-Satoshi-images?type=design&node-id=407%3A1645&mode=design&t=vuHD8KdxXnXQER8K-1)