A GGUF conversion of this great work https://huggingface.co/Phr00t/WAN2.2-14B-Rapid-AllInOne
simple JSON workflow (you need only gguf loader custom node): https://civarchive.com/api/download/models/2099499?type=Training%20Data
umt5xxl and vae excluded. you can use separated ones from wan 2.1
Description
FAQ
Comments (21)
I rarely comment, thank you very much dude!
Could you possibly make a gguf version of v7 which added nsfw loras?
i think usage v1 gguf with separated nsfw loras is more flexible and controllable. maybe i test v7 nsfw model later and check for quality. if this gives good and interesting results i convert this to gguf
DooFY87 v7 is perfect model
DooFY87 Or wait for v8 nsfw - it's in cooking and should be greater.
This Q3_K_M model doesn't work at all, error: "KSamplerAdvanced
Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 68, 5, 100, 68] to have 36 channels, but got 68 channels instead", I use your workflow, changed decoders, NOTHING!
Works fine here, try downloading the file again? I used the OP's workflow and it's working surprisingly well for me, at least in 480x480
It works fine. Something is wrong at your comfyui - maybe update it and custom nodes also. (many times old lora manager custom node make errors for exemple).
FloatsYourStoat Nothing!, it doesn't work, I changed the extension, it still doesn't work..., but other models work fine on the same workflow. Very strange...
flo11ok874 I am constantly updating ComfyUI.
DJKayF And what are you using for the vae? and the clip? I'm personally using Wan_2.1_vae.safetensors and umt5-xxl-encoder-Q8_0.gguf. Do you get any sort of errors when you launch comfy? any missing pip packages for example? Do you have a venv so changes to the system python doesn't cause problems with comfyui? Just throwing out ideas, if everything is in working order it shouldn't be failing to generate.
Wait! hold up, I googled it, someone else had this problem and it was caused by a custom node, do you have something called "flow2-wan-video." if so uninstall that. (edit as a side note, someone said uninstalling it didn't remove the folder from their custom_nodes folder, they had to manually delete it as it was still causing problems)
FloatsYourStoat WOW! WOW! WOW!, you must be a genius, I removed the flow2-wan-video package and everything worked, the generation went. I'm surprised because I thought that this package was needed for Wan video models. Thank you SUPER!
DJKayF Glad that fixed it for you! Happy genning.
You've done an outstanding job. The v7 version now features excellent nsfw content. Quantify this model version, this will be a truly perfect piece of work.Thank you for your selfless work
The v8 version by the original author has been released, addressing noise issues and occasional image blurriness. That's a truly improved version. look forward to your GGUF version，great job!
I'll jump in here to voice my desire for this too. The Q3_K_M version of Rapid WAN here is the only WAN checkpoint my hardware is reasonably capable of running at all, and even then only without loras.
check new rapid v9 gguf version
@DooFY87 Many thanks for that. Author working at v9 nsfw also if you want to see in future and maybe gguf it also.
@DooFY87 Thank you so much!
@DooFY87 thanks a lot :) Are you going to release the new T2V model too?