Confused on how Stable diffusion is able to create image for non-512x512 pixels

[MaxTran96]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What happened?

Hi i'm fairly new to Automatic and Stable Diffusion, from what I read, SD 1.5 only produce output for size 512x512 pixels, but in Automatic1111 UI, it's able to create for 450x800 pixels when i adjusted the width and height. Is this a bug? or a feature? If it's a feature, how does it work when SD only produce 512x512 output

Steps to reproduce the problem

Adjust the width and height, click on Generate

What should have happened?

it should not allow you to generate non 512x512 photo

Commit where the problem happens

idk

What platforms do you use to access UI ?

Other/Cloud

What browsers do you use to access the UI ?

Google Chrome

Command Line Arguments

No response

Additional information, context and logs

No response

[MaxTran96]

Sorry if this is a silly question, i'm just curious on how Automatic1111 able to make SD produce a non 512x512 image output :3

[ClashSAN]

the original compvis repository allowed you to produce sizes multiples of 64.
eg: 512x768 , 512x256 , 512x576

The hack implemented here by @aliencaocao #4978 allows multiples of 8.

NOTE: It works in any resolution that are multiples of 8. If a resolution other than multiples of 8 is given, the generation will succeed but actual generated image will have its resolution rounded down to nearest multiple of 8. In this case, PNGinfo will be INACCURATE as it will falsely document the generated image as having the non-multiple of 8 resolution.

[aliencaocao]

512x512 is just what the model does best as it is trained on images of such size (NAI-related models is 768x768). This is by no means a limitation.

[MaxTran96]

so that means you can make stable diffusion to output image of any size? What about size of not multiple of 8 such as 773x773 which are prime number?

[MaxTran96]

oh nvm just read the comment above. Thanks!