SDXL Resolutions and AI Video
A discussion of standard SDXL resolutions and relevance to making videos
I want to share some experiences with creating AI images and videos in terms of resolutions. I feel I made some mistakes here for myself which is hard to get out of.
Let us first get an overview over recommended Stable Diffusion XL (SDXL) resolutions. They are made so that the number of pixels is alway around 1 megapixel. That is the number of pixels SDXL was trained to handle.
In other words you can use different resolutions from what you see in this list as long as you stay within 1 megapixel.
| Resolution | Aspect Ratio | Common Use |
|---|---|---|
| 1024 × 1024 | 1:1 | Square / general purpose |
| 1152 × 896 | 9:7 | Portraits, close-ups (landscape) |
| 896 × 1152 | 7:9 | Portraits, close-ups (vertical) |
| 1216 × 832 | 19:13 | Editorial / storytelling (landscape) |
| 832 × 1216 | 13:19 | Editorial / storytelling (vertical) |
| 1344 × 768 | 7:4 | SDXL “HD” widescreen |
| 1360 × 768 | ~16:9 | HD Video (non-standard SDXL resolution) |
| 768 × 1344 | 4:7 | Vertical video / social media (SDXL) |
| 1536 × 640 | 12:5 | Panoramic / mobile (landscape) |
| 640 × 1536 | 5:12 | Panoramic / mobile (vertical) |
I really liked the 19:13 aspect ration you get from the standard SDXL resolution 1216x832. It is wide like a typical video 16:9 aspect ratio but offers more height which is good when you have your sex girls you want to show off as much as possible. Sure you can just fit her within the frame, but it means a lot more pixels are spent on the background rather than her.
Making a wide video is not without tradeoffs. The more a character covers the whole screen the more good details you get. Quality of a character degrades quiet a lot if you show them from head to toe in a 16:9 resolution.
I picked the 19:13 a bit by accident, as I wasn't really thinking about video when I made many of my images initially.
