Open utterances-bot opened 1 year ago
This is a great article! A lot of valuable information to serve as guidelines to future steps. Having a passion for both origami and AI, I have also explored (at least conceptually) the possibilities of combining them. I've actually had some ideas regarding the text-to-stick step that you mentioned, however the idea I had was to use an image instead. I think this is actually something simple to do but of course it's the easiest part of the full workflow. Not sure if you saw this, but Tadashi Mori recently made a video about an origami he designed with AI. And this one was actually designed by AI, i.e. the AI made the CP (although there was no prompt inputted). It is a different approach than the one you mention here since he used a Genetic Algorithm to progressively update a CP that is at least flat-foldable. I'm guessing the approach you describe here will be the one used in the future but this is also another interesting use case.
PS: I'm the on who designed the origami for the challenge proposed by Grant Marshall :) I do believe that there is still much to explore with this "inspiration part" before we actually have some AI designing origamis from text prompts since. Until now, people either used a real subject, or their imagination, when designing origamis. Using your imagination can be difficult since you don't have a clear reference (unless you somehow make that concept a reality by drawing it for example) so Midjourney and other tools can be used to bring that concept to an real one, thus facilitating the design process.
@Goncalo-Chambel Thank you for your interesting comment, and congrats on the dragon egg design! I haven't seen Tadashi Mori's video yet, but it sounds very interesting.
Hi Michał, what a great article. I am also an origami fan as well as an AI enthusiast. The idea of using AI applied to origami has been running around in my mind for months now. The idea of generating a tree diagram from a prompt is great and I think relatively 'easy' to implement. The technology is there. We only need to find or 'generate' the training data... I have been playing with the idea of training an AI to recognize origami models in pictures. The idea is generate artificial data using a 3D SW like Blender. You can easily generate several hundred pictures of for example cranes. Then you put the pictures on real photographs and retrain an existing AI on those images. It should be able to identify the cranes in real pictures. I only need the time to work on these ideas. I will keep you posted if I do something
It should complete a full diagram
@fsanchezsb Thanks. This is an interesting idea. Do let me know how it develops.
For the record, here is Tadashi Mori’s video: https://youtu.be/dPPg-s4lK6Q
Origami AI — a roadmap - Origami by Michał Kosmulski
In this article, I present a roadmap for applying AI to the field of origami. Starting from what is possible today, I suggest how existing building blocks could be used and extended to bring us to a point where systems can design origami automatically based on user input. This is both a prediction of where things might go and a to-do list.
https://origami.kosmulski.org/blog/2023-01-12-origami-ai-roadmap