Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add support for generating images with different sizes. #111

Open
ZeroCool940711 opened this issue Jul 14, 2022 · 2 comments
Open

Add support for generating images with different sizes. #111

ZeroCool940711 opened this issue Jul 14, 2022 · 2 comments

Comments

@ZeroCool940711
Copy link

Hi there. It seems like the images generated all have a fixed size of 256x256 pixels not only on the frontend but also when saving them to disk. I think it would be nice if we could have a parameter to specify the size of the images been generated when running Dalle Playground just like how we can set a port or configure the model size, even better would be an option on the frontend/interface from which we could select the image size we want from a drop-down menu or set a custom size, that way we can create images that can be used for wallpapers or for other practical uses, in my case I was able to generate some awesome images that could be used perfectly as wallpaper but because of the small size I can't use them for that.

@youmebangbang
Copy link

The small size is because of the decoder dimensions of the model which has been set to 256x256. Dalle-2 will have 512x512 because the decoder network is larger. This is with VRAM and memory maximized. It is just the current state of AI and in the future you will see much higher resolutions and fidelity.

@ZeroCool940711
Copy link
Author

@youmebangbang I understand, in that case, would it be possible to implement it by doing something like what is mentioned here on this article?
What is done there is expanding the images generated with DALL-E Mini by using VQGAN and CLIP to inpaint the sides, this will create a bigger image and for what I see it can be done to get an image of any aspect ratio and probably of any resolution with a bit of tweaking the code, unfortunately I'm not good enough coding to do that myself. By using the code provided there on that article or here if it is added to Dalle Playground it would be really useful.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants