3 ms·
Yes! A really cool way to work with it is to generate a bunch of images, arrange them on a transparent canvas (i.e. in Affinity Designer), and then ask Dalle to
by cube2222 4y ago
Yes! A really cool way to work with it is to generate a bunch of images, arrange them on a transparent canvas (i.e. in Affinity Designer), and then ask Dalle to fill in the gaps.
For example see here[0], where I've combined a picture of a flying whale, a tardigrade in space, and a bunch of flying turtles.
[0]: https://labs.openai.com/s/quqITCrFI7h0G1HKyfUrmJU0 https://labs.openai.com/s/quqITCrFI7h0G1HKyfUrmJU0
- jordanmorgan10 4y agoThat is neat. So are commas an official way to blend different schools of thought together for the image? Is there any documented way it’s supposed to work? Like [main subject], [art style], etc? Or is it something you picked up from trial and error?
- cube2222 4y agoHere[0] is a very good presentation about it. But overall, you have to think about the context Dalle has seen similar images in the training set. If it's seen them on an art sharing site, then it's probably good to mention such sites and tags it could hypothetically have there. Or if it's more like a photo in an article, think about what could be written about it in the article. That's my intuition about it at least. [0]: http://dallery.gallery/wp-content/uploads/2022/07/The-DALL%C2%B7E-2-prompt-book-v1.02.pdf http://dallery.gallery/wp-content/uploads/2022/07/The-DALL%C...