wan animate contiune motion

[xyzDist]

Hi @kijai
just wondering do we have the function of contiune motion like the native one on "wan video sampler"?
how do we do the same thing on the wrapper?

If we don't have this function, what do you think if it is possible to add? I think this is very useful! since I still have VRAM limit on long duration with using context options, and a lot of benefits split the tasks into different jobs and re-run, or simply to contiune a video.

Thank you very much!!

[xyzDist]

ah. I saw a similiar request from another post
#1533

[railep]

I don‘t know whether kijai has the time or interest to add this but you can easily create a workflow with his nodes which splits a long videos in various parts and stitch it together. Should be consistent (without visible breaks) if you use the same seed.

[RuneGjerde]

Kijai will know better, but i think thats built into the WanVideo Animate Embed, where each window is 77 frames, and continues motion automagically.

And you also have Contect Window generally speaking, but for some workflows this is not needed since the specific node has continued window built in

[xyzDist]

HI @RuneGjerde,
the default wanVideo Animate Embeds have degradation for long duration, result looks bad, so I opt that out.
I am using context options already, but with that still have VRAM limitation, as my machine 4080s 16g VRAM only can go 1000 frames... with 512*1024 for example, thats why I want to split and continue the tasks.

[aidifussion]

HI @RuneGjerde, the default wanVideo Animate Embeds have degradation for long duration, result looks bad, so I opt that out. I am using context options already, but with that still have VRAM limitation, as my machine 4080s 16g VRAM only can go 1000 frames... with 512*1024 for example, thats why I want to split and continue the tasks.

I think this is my same case, like in wan2gp It can continue the previous video motion, but slower than in comfyui..

[kslz]

HI @RuneGjerde, the default wanVideo Animate Embeds have degradation for long duration, result looks bad, so I opt that out. I am using context options already, but with that still have VRAM limitation, as my machine 4080s 16g VRAM only can go 1000 frames... with 512*1024 for example, thats why I want to split and continue the tasks.

Hi, thanks for sharing your experience — that makes sense, especially with the VRAM limitation on 16GB cards.

I’m currently in a very similar situation: I’m already using Context Options, but due to VRAM limits I still need to split long generations into multiple runs. What I’m trying to figure out is how to continue a new run based on the last few frames of the previously generated video.

So I wanted to ask:

Do you know a practical way to pass the last N frames from a previous WanVideo generation into a new workflow, so that the motion / scene can continue more smoothly?

For example, are you reusing the last frames as ref_images, pose_images, or in some other way?

Or is there any method you’ve found to persist or re-inject context between separate runs?

If you have a concrete workflow or node setup for this, I’d really appreciate it. Thanks in advance!

[xyzDist]

Hi @kslz , I think as we now have "SVI 2 pro", it should be the best approach to address the long duration video issue. you can have a check.

[wuwukaka]

I've added a parameter named transition_video in PR #1946. This allows users to pass the last 32 frames of the previous video to achieve a seamless transition between two video segments. It functions similarly to continue_motion, but also supports both context and loop modes.

[xyzDist]

Hi @wuwukaka This is awesome!! is there an example workflow? thanks!!

[wuwukaka]

Hi @wuwukaka This is awesome!! is there an example workflow? thanks!!

You just need to pass the last 32 frames of the previous video into transition_video. I do have a example workflow, but how should I send it to you?

[xyzDist]

Hi @wuwukaka This is awesome!! is there an example workflow? thanks!!

You just need to pass the last 32 frames of the previous video into transition_video. I do have a example workflow, but how should I send it to you?

could you send to my 3donwong@gmail.com please? thank you!

[wuwukaka]

Hi @wuwukaka This is awesome!! is there an example workflow? thanks!!

You just need to pass the last 32 frames of the previous video into transition_video. I do have a example workflow, but how should I send it to you?

could you send to my 3donwong@gmail.com please? thank you!

I have already sent the example workflow to your email.

[zead-ai]

Hello @wuwukaka , can you consider posting your example workflow on https://pastebin.com/ for everybody interested in this example of transition_video workflow ? I am interested too ! thanks !

[wuwukaka]

Hello @wuwukaka , can you consider posting your example workflow on https://pastebin.com/ for everybody interested in this example of transition_video workflow ? I am interested too ! thanks !

I'm not familiar with this website, but I'll give it a try.