Innovate Futures @ Benji
patreon
(For Patreon Supporters) Wan Vace Img2vid With Extend Long Video Mutlitalk Lipsync - Advanced - 20250627
🕑 Added 2025-06-27 13:00:14 +0000 UTC
Tutorial : https://youtu.be/Gj1AwrOmyJ4
Related Post : https://www.patreon.com/posts/132367995
This workflow are basically using Wan Vace StartEnd Frame flexisble functional to extend video length AS MANY AS YOU CAN.
And we generate lipsync for the character after the long length video generated.
Tips when use :
Find out the LightX2V Lora do not works well when you are using FusionX model, T2V I2V or VACE. Color change , Over saturation happened after 3rd sampling. Through my testing examples in Youtube you can see the difference.
Make sure your Audio length are same or short than the video. So your audio can render all duration and use up all available time frames from the video.
Sampling : MultiTalk - CFG :1 , Shift 1 or 2 (can be test depends on video clip)
Denoise , for MultiTalk V2V in the range of 0.6 - 0.7 do the best.
I also attached the other 2 basic workflows, in case you are new to ComfyUI, start from the basic.
Models , Custom Node that I use in this post https://www.patreon.com/posts/132367995
Have Fun!
Comments
Alexis Alvarez
Hi Benji, how are you? I’ve been trying to run this pipeline on my computer, but with my limited resources it’s painfully slow. Which platform or online service would you recommend to run it more smoothly? Thanks a million for your help!
Toon Gaming
Is it normal for the Multi-Talk render to lose some quality? Part 1/Part 2 look A++ Combine A++ but then the multi-talk Combine renders at C+/B- quality. The lip sync seems to work though, I am using LoadAudio>VideoCombine & LoadAudio>AudioSeperation>MultiTalk Wav2Vec. Wonderful workflow, extensions are amazing thanks!! Edit: I must have messed something up when switching the audio nodes around because I have it working now. Obviously User error < .. Highly recommend this workflow, it's going great
Benjamin Law
I guess you are using the WanVideoWrapper main branch , so therefore it doesn't have the Wav2Vecb model need for MultiTalk in that version. Please take a look at the link below. You can use the multitalk branch from Kjai wrapper, or the Forked wrapper_MultiTalk node. Then you will have the Wav2Vec for MultiTalk model. https://www.patreon.com/posts/132088744?utm_campaign=postshare_creator&utm_content=android_share Thanks
b1024
Hi benji, i am fail on (Down)load Wav2Vec Model, please help :( DownloadAndLoadWav2VecModel expected str, bytes or os.PathLike object, not NoneType
Benjamin Law
Installation https://www.patreon.com/posts/132088744?utm_campaign=postshare_creator&utm_content=android_share
hj Kim
I can't use MultiTalk. Current version Python version: 3.12.10 pytorch version: 2.8.0.dev20250609+cu128 Is there a problem? I tried several times but it still doesn't work. Is there a separate video on how to install MultiTalk? It shuts down in wanvideo model loader. Sorry, my friend. My English is not good enough, so I'm using a translator.
