Official ComfyUI workflows with fl2va int8 version of the Minimax H3 model generating 8 sec 720p video on rtx4080 16GB + shared RAM: 32GB
with no optimizations generates for 25 min
with FirstBlockCache reduces generation to 15 min - lossless quality
just works or set mode to H3 Aggressive for even more speed
with Turbo Lora reduces generation to 6 min - with latest version the quality loss is minimal on 6 steps
Other changes from official workflow:
replaced Load Image node with Image Crop for convenience to make the image in the video aspect ratio to prevent stretching
replaced Save Video with VHS version of it to improve the codec quality of the video file
added notification node to alert when generation is finished and display generation time
added Comfy Kitchen Attention node for optimization
Updates:
changed all models to int8 - better quality
updated turbo lora - better quality and speed
Description
After tests updates:
best model versions
best settings
based on minmax speed and quality, trying to avoid limits on rtx 4080 16GB VRAM + shared 32GB RAM
FAQ
Comments (6)
Wow, it's so cool that you're promoting with Wuwa characters 😍 And the workflow is amazing. Thank you for your hard work.
hi. does td that you used in workflow work with 16vram+32ram?. it is about 25gb.
it uses ram as shared vram, and even if the ram is not enough it uses ssd to offload memory because most of the time you don't need the whole thing loaded in the vram, of course it will become slower with the all the memory juggling
text encoder is used only for initial phase of the video generation which handles prompt understanding so it will not make it much slower and you will have better results in the output
@goonjin thank you for answering
which anime to real LoRa are you using?
just chatgpt image edit with simple prompt "make it realistic"