WAN Animate Workflow Updates! (Patreon)

Video - https://youtu.be/rI2lK1GCDc4

Overview

A few updates to the WAN Animate workflow featuring:

WAN Animate v2 Model (smaller file size)

New Face & Pose detection nodes & models

Frame counting :)

Bouncing / Looping videos

I hadn't noticed this before, but since playing a lot more with audio clips it became apparent there were some rather curious frame issues that weren't just related to the looping embeds!

The best input videos have nice frame counts that divide nicely by the chosen frame window size.... if not, then extra frame will be added.

One option is simply to use the "trim to audio length" option in the Video Combine node, and for most cases that will do just fine!

However... other nodes can also cause issues, so I figured I'd share all my findings!

Frame Window Size:

By default the VID_FRAMES count is connected to the frame_window_size setting because that will make sure the generated video is of the same length as the input video. This is different to the default size of 77, but you can simply disconnect it to manually set the size.

Larger window sizes may not match the reference image as well, but generally speaking this effect can be countered via more accurate prompting that matches the reference image.

It works well (so far as I can tell) with masked video, but for "image animate" it can mess up the first frame.

Load Video Node

Depending on the Load Video node you use, you may get a frame count which actually DIFFERS from the one shown in the node! The FFMPEG version of Load Video tends to be:

More Accurate

Better quality

I found this out by adding a few frame counters, which you can test and check as well!

Audio Clipping

I also noticed that some audio would be clipped as well - even if the input and output video both appeared to have identical frames.

The workaround in this case? Us the NATIVE ComfyUI Load Video node, using only the AUDIO from that. For whatever reason the images produced by the native node seem to produce errors with Kijai's nodes :/

New Pose / Face Detection

I put these models & nodes through their paces with some rather complicated scenes, and for the most part they held up pretty well! I've only tested the large model, but a huge model option is available as well.

Complex scenes work ok with the masked video option, but for image animation things may not work out quite as well...

Attachment Info

Top-level locations for Kijai's collection files:

https://huggingface.co/Kijai/WanVideo_comfy/

https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled

See named sub-directories for various file options, e.g.

https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v

https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_relight

Multiple file options are available & remember that you don't have to use the workflow defaults. Experiment, learn, download the others and have fun with your creative explorations!

models/diffusion_models:

V2 Example: https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/blob/main/Wan22Animate/Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors

(Last updated 07/Oct/2025)

Detection Models

Download to models/detection

NOTE: You will need to make this directory yourself, and this location is subject to potential change in the future!

-- YOLO:

https://huggingface.co/Wan-AI/Wan2.2-Animate-14B/blob/main/process_checkpoint/det/yolov10m.onnx

-- ViTPose Large:

https://huggingface.co/JunkyByte/easy_ViTPose/blob/main/onnx/wholebody/vitpose-l-wholebody.onnx

Custom nodes can be installed using the usual git clone command, or via ComfyUI Manager.