Replies: 4 comments 2 replies
|
Looks about right to me. What I do instead is use the fastwan finetune. It's not only faster (distilled) but also looks way better. https://huggingface.co/Green-Sky/FastWan2.2-TI2V-5B-FullAttn-GGUF also feel free to try the tiny vae decoder to save vram #937
Not used by 5B. Also try |
|
Or try Here is what I can manage on 8gig vram in 30sec:
fastwan2.2-t2i5b_tae_3steps_euler+lcm.mp4 |
|
Thank you both of you. Now seems it starts to work. |
|
I have plan to move into img2img generation that's why i use dual model. I wonder is there some Wan 2.2 that makes even better quality, but in same time fits into my 12GB VRAM mostly(rtx 4070). I would not mind double encode time, juts not a slow as snail. if there is some i would be grateful for exact model combination that do not fail. Sys ram is 64GB. I am beginner in this local AI stuff, but i chose stable-diffusion.cpp as it is better and more compact and lighter and .cmd scripts are good for me personally. I do not like those "fat" Python programs. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Is it my mistake or something other wrong, i even cranked steps up and added extra options, but still garbage video comes out. Text to image mode tested, i have not tested img-to-video mode.
test output.zip
WAN22 cmd.zip
All reactions