Opening here for tracking/awareness. Kos wrote this on Discord:
ref cn will not work with tiled diffusion at this time
the cn would need to have access to a vae during sampling to encode the input image per tile, and currently it prepares them beforehand. i'll eventually refactor code to make it compatible with existing tiling solutions, but for the time being i don't want to introduce potential memory usage issues without putting in the proper time to investigate
Opening here for tracking/awareness. Kos wrote this on Discord:
See also https://github.com/shiimizu/ComfyUI-TiledDiffusion/issues/20