Do you need to subscribe paid plan in openai to generate code? wont the free one do?
don't think so. from what I see you need to put at least $5 into it
Do not subscribe to ChatGPT Plus subscription for Abyss Diver! The subscription is not necessary for the avatar generation and is a completely separate cost!
What you need to do is buy $5 worth if credit on the API platform, and then you can use the feature in the game. Theoretically you should be able to use the free credit, but from what I’ve heard that disabled free access to DALL-E 3 for some reason, even though their website says it should be available. ♀
Also, in a new update that should be coming very soon, I’ll be giving the player the text of the prompt I’m using for the generations, so you can use that in ChatGPT or whatever other software you prefer.
Wouldnt it be more sensible to grab the images from a local StableDiffusion installation? You can get it from github and generate the images on your device with training packages you can grab from other sites for free aswell, tho I dont know how much work integrating that is
So, I’m a big fan of Stable Diffusion, and I actually used it to generate many of the images in the game, but unfortunately it doesn’t work well for this. Stable Diffusion often relies on optimization of the prompt, and even then it often misses details and doesn’t really understand what the intention of the image is. Whereas DALL-E has a few features that are great. The first is that the API has an integrated language model that optimizes the prompts automatically, given some basic things it needs to include, which removes the need to have a person optimize the prompt. And also the understanding of the prompt is so much better, so it’s much more likely to get details, which is very important when I’m giving the model a bunch of physical characteristics to include and I want all of them to be in the final portrait.
And of course there is the other issue brought up, which is that doing it locally is actually a lot of problems. I would need to distribute the diffusion model weights, which would be 5-10 GB on their own! And then players would need to get the model running on their computers, which would be a major pain. And it would be impossible on mobile lol.
I would love to eventually move to an open model, with no censorship (for obvious reasons for this game lol), and that players can run for free themselves! But
unfortunately I don’t think that’s viable yet for a variety of reasons. I appreciate the suggestion though!