Generate High Resolution Image with Stable Diffusion WebUI

Stable Diffusion models are typically trained on 512x512 images. If you try to generate larger images of people, you may end up with two heads or weird limbs.

Fortunately, Stable Diffusion Web UI has a Hires. fix feature to resolve this issue. Here's how to use it:

• Set your image size to a size based on 512, for example, 512x768.
• Check Hires. fix.
• Choose an upscaler, e.g. 4k-UltraSharp. Upscaler works differently in Hires. fix compared to extras. In Hires. fix, you will get a native high-resolution image, while in extras, it only performs super-resolution upscaling.
• Set the number of Scaled by, e.g. 2, you will actually get a 512x1536 image.
• Last and most importantly, set Denoising strength to a value below 0.5, otherwise you may still see two heads or weird limbs.