Update version +color
Color fix at the end of each frame-bucket (be carefoul - could destroy image)
Change FastWan -> Fussion
Step downgrade 6 -> 4 (yes with Fussion 4 is really beter that 6)
Update version +clip
Added vision_clip on each first frame, that fix morfing a lot.
Ordinary you can't create video more than 81 frame (you can, but context lost very fast)
Visible way to fix it - create 81, take last frame and create next 81
This workflow contain For-Loop so you can create as much step as you wish. In terms of usable:
Good idea for prompt: "Girl danse and undress", "Girl flirting and smile poses on camera", "Girl masturbate and feel pleasure"
Bad idea: "Car crashing on wall", "Apple drop on the floor", "Girl has a squirt orgasm"
In other ward do not use "finished action" in your prompt.
Description
Change default lora settings, add color correction node.
FAQ
Comments (9)
Love the workflow. Unfortunately the color fix is not working well enough. I was using that in 2.1 already I really hope someone comes up with a solution somehow. <3
Yap! It's really hard to understand what part is in the frame and what is gone.
thesharque588 I works well enough for 2*5 or maybe even 3*5 second for me, but then the quality gets real/too bad. Not even sure this can be really fixed with Wan2.2.
bowiba1265909 It coms from 2 main problem.
1 - Even in "first -> last" WanFunFlow Last frame != original frame. It's recreated by WAN model and complettly different. (to make sure just take a photo from internet and downscale it, U'll C that image ruined after 2-3 loop
2 - Wan context < 81 frame is good, but after 81 frame it's jut "don't know what to do..." because of learn process (it's based on short samples"
Today I plan to publish new workflow (stupid and simple) but it should help to resolve this shity problem : )))
thesharque588 I am just a user but I figured that is/should be the problem. I am thankful for any nice workflows shared! Looking forward to it.
bowiba1265909 It is a problem (Actually it is a huge PITA and "known issue" : ))) but there is no perfect solution for now.
@thesharque588 I was wondering if there is some kind of node that allows to switch the input image?! I mean run one input image X times a 5sec. I usually do 3*5 because after the quality gets too bad. But the next run I want to use same prompt but upload new image for starting point. That way I could run multiple over night. Do you know by chance? I could not find any, yet.
bowiba1265909 if you want to have completely different frame just use copy of workflow
thesharque588 I wanted to up the batch count of comfyUI to 100 but get a new starting image every time