A Minor Issue with the Combined Use of Lora-Block-Weight and Multidiffusion-Upscaler-for-Automatic1111

[YondSun]

When I use Lora-Block-Weight and Multidiffusion-Upscaler-for-Automatic1111 simultaneously, activating tiled diffusion with controlled partition prompts, and after writing prompts like lora:abc:0.8:lbw=FACE in specific partitions.

If I choose the DPM++ 2s a Karras sampler, the generation speed increases by 2 ~ 2.5 times.

However, with the DPM++ 2M SDE Heun Karras, the speed remains normal.

This issue occurs under quite specific conditions and may not be universally applicable, but I still want to raise an issue to see if any help is possible.

[hako-mikan]

That issue is deep-rooted, so it will be difficult to address. Probably, it will have to be fixed starting from the sampler.