Location: eastern US Interests: dogs & other cute things & people. AI & pixel art. AI for writing. And I have a bunch of kinks mostly involving transformations of various kinds & gender swapping & nullification. Pronouns: she/her.

  • 39 Posts
  • 24 Comments
Joined 1 year ago
cake
Cake day: July 24th, 2023

help-circle

  • I’d heard it was harder, but that may be with the base model. It’s certainly been much easier so far when using SDXL Unstable Diffusers ☛ YamerMIX ! It doesn’t work with tools like ControlNet, though, so it’s limited to prompting and img2img. It might work for some limited use cases in Deforum that don’t require ControlNet, like zooming with high strength. It’s hard to get good videos in Deforum without zooming, though. (I’m trying to make videos with Deforum now.)














  • Rendering at larger sizes can also make it more likely SD will generate multiple subjects. You can use the Regional Prompter or Latent Couple extensions for stable-diffusion-webui to control the subjects separately.

    You can also generate subjects separately using txt2img, paste them into a single scene, then do a pass with img2img to make the image look seamless. That is about the hardest approach, though. ControlNet with OpenPose and region prompter are my go-to solution.