Basically, it generates in low resolution first and then finishes at the high resolution. It uses just the basic latent upscale, so it's a bit unstable—but hey, it's just for my own use anyway.
I have only tried T2V and I2V. There might be an issue with the reference functionality. There is a problem with my stupid implementation. If a problem occurs, try replacing the Latent Upscale part with this: https://github.com/Tr1dae/ComfyUI-MiniMaxH3_LatentUpscaler
