This workflow allows you to use a motion path to move your subject or camera using Vace+Wan2.1
If you need more info on how the workflow works, check out the link below:
Description
FAQ
Comments (14)
This is amazing. Is it possible to have multiple paths? For instance, in your dog example, I wanted to draw two lines: one for the dog, and one going the opposite direction for the background, so that it would know to make the dog move away from the direction of the background, while following the dog a little with the camera.
Same goes for making the man's hand stay steady, while moving the bird. Even if it's in slow motion, it seems it might understand to make the bird move separately from the hand.
So, would it be possible to overlay the different boxes using some kind of image overlay node, and have multiple boxes going at once? Will WAN understand this?
I found a way to add more splines . you need to use AppendStringsToList kjnodes. you pull in bothe spline a and b into this node and pull the output to the cordinates input on the CreateShapeImageOnPath.
you need to make 1 extra spline node and hook up frames width height and the bg image as the one thats already there
hope this helps may make a new workflow using the fun14bwan21 model instead but with multiple inputs
you can daisy chaine the AppendStringsToList so you can add as many splines as you want
@carlosgms81410 do motion paths work with wanfun too?? Hadn’t tried that!
@theartofficialtrainer they do but you need specific color setup for the motion shapes :
Black BG
Yellow outlines 10 pixles thick
Red infill for shape
and i use a size of 20 height x 50 width
it does work but there are some issue as any part that dosent have a path dosent get updated so will probably need a 2 pass aproche 1 to give the main motion and 1 to cleanup the generation
@theartofficialtrainer hmm my replay disapered XD ,
They do but they need a specific color for the spline shapes :
Size: 20 height X 50 width
BG color : black
Shape color : Red
outline color : Yellow
there are some limitations , it does not animate anything without a path so this will probably require a multiple approach to make it viable : 1 pass to get the main motion and one to clean up the video with fewer steps and denoise in lower value , when i get it to work ill post a workflow in my profile .
fyi i got multi spline control working can add as many as i want :D
idk what the heck im doing wrong. im trying to run this local. ive downloaded all the same models, watched your video. im not running docker, etc. i also installed sageattn and all i get is RGB noise and a broken result.
and no matter what frame count i set it to, it always does 85
@Knoxxxonk made a coment to the author as well but here you go :
there is a small issue for me the set count node is missing and the "WanVideo VACE Encode" sets the wrong count of frames , changing the widget back to input shows 85 instead of the 81, can be fixed by manually inputting frame count or setting a SET Frame count equal to frame count image load
so there is a missing node so just right click the WanVideo VACE Encode and click the convert input to widget num_frames and you can set the frame count manual on this node
also the garbled RGB noise sounds to be wrong model loaded check the VAE load , Text encoder , and the actual model loaded , there are 2 model loaded VACE 1.3b and wan21 1.3b t2v
hope this helps good luck
@carlosgms81410 thanks i was able to fix the frame count!
as for the models, i believe i downloaded the same exact ones. i have the following
WanVideo VACE Model Select -> Wan2_1_VACE_1_3B_preview_bf16.safetensors
WanVideo Model Loader -> Wan2_1-T2V-1_3B_bf16.safetensors
Load WanVideo T5 TextEncoder -> umt5-xxl-enc-bf16.safetensors
WanVideo VAE Loader -> Wan2_1_VAE_fp32.safetensors
i believe this should all match to what the default is, only difference is i dont have them in /wan folders but that shouldnt matter
i also changed a few things but i dont think it should be the issues. i had to bypass the create shape on path node and its preview, it was not liking the values in there. i tried editing but i think its just a preview anyways. and even though i installed sage attn it seems today its bugging out when trying to run with it so i switched to sdpa.
@Knoxxxonk
yeah you seam to have everything correct. if you have issues with sageattention just use SDPA as its only a 10% difference in speed no quality lose .
does it render correctly with SDPA now ?
@carlosgms81410 no unfortunately its still just garbled RGB noise besides the first frame
there is a small issue for me the set count node is missing and the "WanVideo VACE Encode" sets the wrong count of frames , changing the widget back to input shows 85 instead of the 81, can be fixed by manually inputting frame count or setting a SET Frame count equal to frame count image load
Great workflow easy to use and decent results
I almost forgot about him with the previous tool for moving points to generate dynamics, I like the vace model the most, the possibilities are endless, thank you
That one dude in the example needs to man up