I need help from wiz, It took me 1 hour or 2 hour to generate 6 sec clips.

Im using RTX3060 12gb vram
is there any other way to speed up the generations ?

you running at local?

>>19775084
yes locally

0. install sage and triton (this is pain todo, https://www.patreon.com/posts/easy-guide-sage-124253103 work well on clean comfyui portable installation)
1. use some GGUF models of 480p (q8 or q6 @ https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf)
2. use some other lora loader like rgthree power lora loader)
3. Add teacache node (just before sampler)....alternative there are some good speed up loras, but i assume you new, so stick with teaC (start at 0.2 & stop at 0.9). this basicly doubles your speed.
4. You don't need to load clip into loras, just into prompt
5. lower your steps, you can get pretty good results with 15 steps

>>19775127
you're a lifesaver!! from 1 or 2 hours to 16 minutes only! im gonna try 25 steps later and see how long will it take

thank you so much!

>>19775793
well thats pretty good boost up. there are loras like causvid or lightx that lets you get pretty decent results on just 4-6 steps

Also note that everytime you change lora settings (or on first run after launch), even slightly, it needs to rebuild and load models again, this takes time (amount of RAM matter at this point)

>>19776243
yes i noticed that after my second run, also does adding more lora takes more time ?

Care to share your workflow json file?

>>19779335
Im rebuilding mine atm, trying to get speed-lorars to work better.....if i get it working somewhat, maybe then

...but the thing is i'm not too keen to reply to posts asking why they get it working =)....there are loads of stuff'n'nodes to hunt down