Textual inversion for 2:1 aspect ratio?

[wedgeewoo]

I am just wondering if the 1:1 aspect ratio is required for training a model. I would like to train a model with images of panoramic photos but they are typically 2:1 Ratio, is this possible or will this cause some issues since the sd ckpt was trained with the 512x512 dimensions?

[cgessai]

I'm very ignorant on this process but I just happened to watch the following video earlier, where the host talks for a bit about doing textual inversion with non 1:1 ratio images: https://youtu.be/7Lxdk89W2K0?t=122 The "Split into Two" checkbox option in the "Preprocess Images" tab/window in the UI. You can click on the "..." (3 dots) in the lower right hand corner of the video, select "Show Transcript" and CTRL-F to find other mentions. IIRC, he speaks about it over the course of the next minute or two.

It has been mentioned in almost every technical source I've read that SD "thinks" in squares but this splitting technique may be able to work around that somewhat. Cheers!