I’m currently building an AI product using Dalle 2 to read a prompt and to create an image based on that prompt!
I’m looking to train Dalle 2 so that the images it’s producing get better and more accurate.
I saw some things about using the fine tune option to train a model but I didn’t see Dalle 2 as an option as a model that can be trained!
Does anyone have any resources on how I can achieve this?
DALLE2 cannot be fine-tuned at this time. I believe Stable Diffusion allows you to do it, though.
Welcome to the forum!
Thanks Paul! Is Stable Diffusion in the docs already?
Stable Diffusion is another model that has nothing to do with OpenAI.
I think part of the problem is safety. Stable Diffusion will let you train with anything, even copyrighted photos/art. I don’t think there’s a good way to stop fine-tuning on commercial art where you don’t have the license, so I’m not sure if OpenAI will ever allow fine-tuning.
All that said, this is one of the reasons I’m sticking with DALLE2 and not using SD (which can do better D&D art sometimes) … because of the way the models were trained…
Hope this helps. AI art is moving so fast, it’s hard to keep up sometimes.
Appreciate you breaking this down I’m a newbie trying to catch up on everything lol
No worries. Happy to help how I can.
It’s definitely not easy keeping up to date!
I’ll try to make more time to share helpful / interesting image AI articles in the “In the News” section.