Innovate Futures @ Benji
patreon
(For Patreon Supporter)(ComfyUI) Wan Vace Native V2V CN Extend Long Video Length Workflow Ver.20250706 (updated)
🕑 Added 2025-07-04 13:00:20 +0000 UTCThe first version of this workflow I have posted last month https://www.patreon.com/posts/wan-vace-native-131544550
Then some people might not understand how to run it.
And I did some update and it became something like what you see in the tutorial video.
So I did a tutorial today in here: https://www.patreon.com/posts/133333679
The workflow I did in this post for you guys Patreon supporters, I extend the length to 30 seconds, 6 sampling , basically longer duration on this one.
Some tips to run this (updated-20250706):
Do your math of the video length before run. Index out range error cause by the number of frame for sampling are larger then the number of frames from your reference video.
For example , if your video is 15 seconds, you don't need to enable all Parts of sampling groups. You can use Part 1 - 3 sampling, each sampler work on 129 frames, 24FPS.Do not use lightx2v lora or self forcing lora with FusionX, and think this is going to be faster. It will turn out the over Saturation, or Contrast change after Part 1 sampling. Example like this test video : https://www.youtube.com/shorts/8SGeSFRhmzc
Apply the right way example : https://www.youtube.com/shorts/tTeOw_pXozE
Everything explain in the tutorial video, and you will need the follow tools...
Custom Nodes You need :
For Tool nodes and Wan
https://github.com/kijai/ComfyUI-WanVideoWrapper
https://github.com/kijai/ComfyUI-KJNodes
https://github.com/Fannovel16/comfyui_controlnet_aux
https://github.com/pythongosssss/ComfyUI-Custom-Scripts
https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite
https://github.com/chflame163/ComfyUI_LayerStyle
For Low VRAM GGUF
https://github.com/city96/ComfyUI-GGUF
https://github.com/neuratech-ai/ComfyUI-MultiGPU
Models (updated-20250706)
--Text Encoder--
(https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/text_encoders)
--VAE--
(https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/vae)
FusionX Wan VACE Model (Only for running with FusionX without lightx2v_cfg_step_distill_lora) :
Safetensors
https://huggingface.co/QuantStack/Wan2.1_T2V_14B_FusionX_VACE
GGUF
https://huggingface.co/QuantStack/Wan2.1-14B-T2V-FusionX-VACE-GGUF
Models (for running without FusionX-CausVid )
Wan2.1_vace
(https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/diffusion_models)
Or
Wan2.1_T2V_14B_LightX2V_StepCfgDistill_VACE-GGUF
(https://huggingface.co/QuantStack/Wan2.1_T2V_14B_LightX2V_StepCfgDistill_VACE-GGUF)
Wan21_T2V_14B_MoviiGen_lora_rank32_fp16
Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32
Wan2.1-Fun-14B-InP-MPS
DetailEnhancerV1.safetensors
(https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX/tree/main/OtherLoRa's)
If you haven't install some nodes, fix it before load the workflow.
Then you can load the workflow and seeing something like this.
Follow the way I did the setting in the tutorial video, and have fun :)
Comments
Light Scheme
Hi Benji, I'm trying to use this workflow for long video to video generation but i'm getting the error "The size of tensor a (130696) must match the size of tensor b (134946) at non-singleton dimension 1" in the first KSampler, would you know why this may be, what i may be doing wrong, are there requirements for exact dimension matches between the source video and reference image? please let me know if you have any ideas how to fix this :)
Benjamin
try GGUF model, lower resolution ?
Paciku123
I run out of GPU memory with this workflow, but I have RTX 5090 with 32gb VRAM. How to fix this?
Benjamin Law
I never use Mac to run AI stuff. I started from nv 4090 and recently upgrade to Rtx pro 6000.
Alexis Alvarez
Benji, is it possible to run this on a Mac? I have an Apple M4 Max, but the multitalk workflow didn’t work for me—I’m stuck at the “WanVideo Sampler” node.
Benjamin Law
__version__ ? what is the ComfyUI version you are using?
Александр Свириденко
Hello. I have a problem with launching ComfyUI portable. I get an error (in the console) when launching: ImportError: cannot import name ‘__version__’ from ‘aiohttp’ (unknown location). I understood that __version__ is removed in the current version of aiohttp. How did you manage to solve the problem?
