Skip to playerSkip to main content
Unlock precise character motion tracking for AI animation! In this step-by-step tutorial, I show you how to take a real-world source video and convert it into a clean OpenPose keyframe skeletal animation to drive generation pipelines like Animate Anyone or AnimatedDiff.

Using a streamlined ComfyUI workflow, I explain how to handle heavy video files by tweaking parameters such as Frame Load Cap, Skip First Frames, and Select Every Nth Frame to break long videos down into stable, error-free chunks. You'll see how to leverage the advanced DWPose Estimator pre-processor node alongside specialized tracking models (YOLOv8 and DWPose weights) to capture faces and body poses perfectly. Finally, we configure the Video Helper Suite's Video Combine node to render the tracking data straight out into a crisp 24 FPS MP4.

Original YouTube Tutorial: https://youtu.be/T9YNARKEkq8

Connected Videos in this Series:
* Video-to-Video AI Animation with AnimatedDiff: https://youtu.be/R2eysgnhmWE
* How to Install & Use ComfyUI Manager (Fix Missing Nodes): https://youtu.be/dyrhPVRsy9w
* How to Install Animate Anyone: https://youtu.be/HBRPA1TsFGQ
* How to Install ComfyUI: https://youtu.be/ynfNJEtvUtQ
* Video Helper Suite: https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite

Source/Model Resources:
* ComfyUI Auxiliary Preprocessors (DWPose Weights Source): https://github.com/Fannovel16/comfyui_controlnet_aux
* Video Helper Suite Custom Nodes: https://github.com/Kosinkadink/ComfyUI-Video-Helper-Suite
* Workflow: https://www.patreon.com/posts/159535844

Video Details:
* Original Publish Date: July 12, 2024
* Focus: Video-to-OpenPose / DWPIes / Motion Tracking / Stable Diffusion Animation
* Hardware Setup: Optimized for local workflows on NVIDIA RTX GPUs (e.g., RTX 4060 Ti)

Follow lordcaocao2025 on Dailymotion for more zero-fluff machine learning guides, advanced animation pipelines, and ComfyUI node optimization techniques!

---
Connect with me:
πŸ“Ί YouTube: https://www.youtube.com/@CaoCao2025
πŸ“± TikTok: https://www.tiktok.com/@caocao20250
πŸ’Ž Patreon: https://www.patreon.com/cw/Caocao2025

#ComfyUI #OpenPose #AnimateAnyone #AnimatedDiff #DWPose #AIAnimation #VideoHelperSuite #lordcaocao2025
Transcript
00:00Hello guys, welcome back with me, Chow Chow 2025.
00:06Today we're going to learn how to create Open Pose video that to be used on Animate
00:11Anyone video, like we share on previous video.
00:15What you need is Comfy UI and Comfy Manager.
00:20Having Comfy UI Manager will help you download any missing node from my very short workflow.
00:28First, what you need to do is open Comfy.
00:31Then you could download my workflow that I upload in ComfyWorkflows.com.
00:37Link on the description.
00:38This node has a few parameter that I will explain.
00:42Video parameter.
00:44This is the video file name that are loaded.
00:47It could be a link if you use video path or it would be file a name if it is
00:51uploaded.
00:52Forced Rate Parameter.
00:54This will adjust your frame rate by discarding or duplicating frames to reach the frame rate
01:00target.
01:02Disable this by setting it to 0.
01:04Forced Size Parameter.
01:07This parameter allows for quick resizing to a number of suggested sizes.
01:13Just pick the option from the list.
01:15Frame Load Cap Parameter.
01:18It set up the maximum number of frames which will be returned.
01:22Skip First Frames Parameter.
01:24It decide how many frames will be skipped from the start of the video after adjusting for
01:29a forced frame rate.
01:31Combining Skip First Frame and Frame Load Cap can help make a long video into a short chunk.
01:38This will help reduce the chance of error happening.
01:42You could always combine the chunk later with other work flow of with software like Avidimo.
01:49Select Every NTH Parameter.
01:53This will allows for skipping a number of frames without considering the base frame rate or
02:00risking frame duplication.
02:02This help reduce frame rate and in the end will also reduce the time to process the video.
02:08Now we will upload a dance video that I created.
02:13Next we will use Control Node Pre-Processors.
02:17Use Add Node.
02:18Then select Control Node Pre-Processors.
02:22Select Faces and Poses Estimator.
02:26Lastly select DW Pose Estimator.
02:30Option here is pretty self-explanatory.
02:34For B Box Detector I use Yolox L.torchgrip.pt.
02:40For Pose Estimator I use DWII Yukoko 384 Bay5.torchgrip.pt.
02:51Last node we use Video Helper Suite Combine Video.
02:56We change the frame rate to 24.
02:59The same with the original video.
03:02Format we change into MP4.
03:05You could select whatever M4 you want.
03:08Save Output Change into True so it will save in ComfyOE Output Directory.
03:14File Name will use FreeFixFollow with Number as a file name.
03:20Now let's run this!
03:25Here are the result.
03:33Here are the result between original video and the Open Pose video.
03:38That's it, guys how to create an Open Pose video from RealVideo.
Comments
lordcaocao2025
Creator
Generating OpenPose skeletons natively inside ComfyUI using DWPose is way more accurate than older modelsβ€”it catches subtle finger and face movements flawlessly. Breaking your source clip down with frame caps is a lifesaver for avoiding out-of-memory errors on long sequences. Did your tracking map catch the hands accurately without jittering? Let's discuss below!

Recommended