>>44763208You need to use Latent Couple and Composable Lora and place the subject on a smaller part of the screen. In my case, the character goes only in the shown portion of the whole image (the preview can't show wide versions, but it's just a strip from a wider image).
You can view how my Latent couple prompt looks like if you PNG info this pic:
https://files.catbox.moe/vp19ak.jpgAfter the image is generated at the highest resolution possible for you,(I can do about 1500x576 before I get OOM, but using 1376x576 for quicker generation speed) you send it to img2img. Use the same exact settings and seed, denoising strength 0.3-0.4, activate Latent Couple and Composable Loras, activate SD upscale script. Here I added a downloaded upscaler called 4x NMKD YandereNeoXL:
https://icedrive.net/s/f0UAiRqz3N (just add it to sd/models/ESRGAN/ folder and restart your webUI for it to be listed) and I am using about 2.5x upscale on my pic, resulting in a 3440x1440 final image.
