Experimental curve compatibility lora node and accelerated Open-H3 Regenerate Fast VAE decode (StreamRender-H3) node
I've been playing around with accelerating the Open-H3 Regenerate fast VAE Decode (StreamRender-H3) node. I've managed to get a 3x improvement in speed using RAM-Staging and lowered VRAM from over 14GB to 8GB and under. I've also created (cough) a curve compatibility lora that gets around the Curve_A/Curve_B warnings seen using the standard Comfy loras with models like Singularity.
For more information please check my repository at: https://github.com/deadman3000/H3Reg2KVAEPlus/tree/main
Hope you find this useful.
Hi Deadman3000,
thank you for your work !
A question please : where can I find the regen_core lib ?
Indeed I have this error :
ImportError: cannot import name 'regen_core' from 'I:\ComfyUI_windows_portable\ComfyUI\custom_nodes\H3Reg2KVAEPlus'
Did you install the Minimax H3 Regenerate 2K custom node through ComfyUI Manager first? You then simply overwrite the two files in that folder with those in my repository and start comfy. I need to make an install guide as they are also loaded as two seperate nodes 'Open-H3 Regenerate 2K - SHarp Curves Only Branch' and 'Open-H3 Regenerate 2k - Fast VAE Decode (RAM-Staged, Experimental)'.