<div data-youtube-video><iframe src="https://www.youtube.com/embed/H5L5sO7dRu8" width="640" height="360" allowfullscreen="true"></iframe></div>
https://youtu.be/H5L5sO7dRu8 
 
Creator: AI-KSK · Collection: Video 327 / six-workflow MiniMax H3 RefMod episode 
 
Workflow file: <code>02_动作RefMod_单视频与文件夹批量.json</code> · Cover file: <code>武术动作参考-2.png</code> 
 
<h3>Main features:</h3> 
 
This graph prepares a motion-oriented RefMod from a source action clip, with the intent of transferring movement cues to another subject in a later H3 video. It is suited to exploring guard positions, step transitions, body rotation, a punch or a kick while avoiding the need to keep reprocessing the same source video. It is not a skeleton rig, motion-capture retargeter or independently trained action LoRA. 
 
The connected chain uses a single <code>VHS_LoadVideo</code> → the <code>refs_video.ref_video_0</code> socket on <code>MiniMaxH3RefModExtract</code> → <code>MiniMaxH3RefModInspect</code> → preview. Although the workflow title mentions folder batch processing and the tutorial describes that option, the actual JSON has no folder batch loader connected. It is a single-video path as shipped. 
 
<h3>Exact values from this JSON</h3> 
 
- Original selected video: <code>203719-922186723_small.mp4</code> — not distributed. 
- Video decoder: <code>force_rate=0</code>, <code>custom_width=0</code>, <code>custom_height=0</code>, <code>frame_load_cap=121</code>, <code>skip_first_frames=0</code>, <code>select_every_nth=1</code>. 
- Extractor: <code>name=Action_01</code>, <code>mode=Full Reference</code>, <code>concept_type=pose_motion</code>, <code>ref_resolution=512</code>, <code>latent_frames=56</code>. 
- Other options: <code>pool_h=16</code>, <code>pool_w=16</code>, <code>motion_only=false</code>, <code>save=true</code>, <code>subfolder=motion</code>. 
 
<h3>Suggested workflow:</h3> 
 
Upload your own usable action footage and select it in <code>VHS_LoadVideo</code>. Check that the first 121 loaded frames adequately cover the motion you want. The 56 latent frames setting is separate from the raw-video frame cap; do not confuse either with a guaranteed clip duration. Run the extractor and inspect <code>motion/Action_01</code> under <code>models/refmods/</code>. If static appearance from the source video is transferred too strongly, experiment with <code>motion_only</code>—but the supplied JSON leaves it off. Later choose this saved motion asset in the Apply or Text Encode generation path. 
 
<h3>Limits</h3> 
 
Reference-driven motion cannot ensure precise joint positions, actor-actor contact or a clean separation of the original performer's clothing from the motion. The action-filmstrip cover is artwork explaining the task, not a produced frame sequence or tested result. 
 
<h3>Dependencies, tutorial and deployment</h3> 
 
This project uses ComfyUI, MiniMax H3 weights and the <a href="https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod" rel="ugc">upstream RefMod node pack</a>; workflows involving <code>VHS_</code> nodes require <a href="https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite" rel="ugc">Video Helper Suite</a>. See <a href="https://huggingface.co/Comfy-Org/MiniMax-H3" rel="ugc">Comfy-Org's H3 model inventory</a> for official distribution guidance. The exact int8-convrot video VAE and Turbo 4-step LoRA named in the creator's JSON need separate availability/provenance checks. We distribute this workflow JSON and its original cover, not external checkpoints or third-party media. Respect applicable rights and licenses. 
 
<a href="https://www.compshare.cn/images/323moGjpvfeM?referral_code=A7fTQUYY9lCDFJwA6MMvoq" rel="ugc">Deploy AI-KSK's H3｜RefMod免训练超级参考 image on 优云智算</a>. Referral disclosure: the deployment link includes an AI-KSK referral code and may support the creator. Check the cloud provider's actual billing and environment before deploying. 
 
<a href="https://youtu.be/H5L5sO7dRu8" rel="ugc">Watch the complete six-workflow YouTube tutorial</a>. This project is one module in that episode. Description and settings were prepared by inspecting the source JSON and transcript, not from a new measured benchmark. 
 
Bilibili tutorial: https://www.bilibili.com/video/BV1F2nZ6SEA7/

MiniMax H3 RefMod Motion Reference

<div data-youtube-video></div>
https://youtu.be/H5L5sO7dRu8 
 
Creator: AI-KSK · Collection: Video 327 / six-workflow MiniMax H3 RefMod episode 
 
Workflow file: <code>02_动作RefMod_单视频与文件夹批量.json</code> · Cover file: <code>武术动作参考-2.png</code> 
 
Main features: 
 
This graph prepares a motion-oriented RefMod from a source action clip, with the intent of transferring movement cues to another subject in a later H3 video. It is suited to exploring guard positions, step transitions, body rotation, a punch or a kick while avoiding the need to keep reprocessing the same source video. It is not a skeleton rig, motion-capture retargeter or independently trained action LoRA. 
 
The connected chain uses a single <code>VHS_LoadVideo</code> → the <code>refs_video.ref_video_0</code> socket on <code>MiniMaxH3RefModExtract</code> → <code>MiniMaxH3RefModInspect</code> → preview. Although the workflow title mentions folder batch processing and the tutorial describes that option, the actual JSON has no folder batch loader connected. It is a single-video path as shipped. 
 
Exact values from this JSON 
 
- Original selected video: <code>203719-922186723_small.mp4</code> — not distributed. 
- Video decoder: <code>force_rate=0</code>, <code>custom_width=0</code>, <code>custom_height=0</code>, <code>frame_load_cap=121</code>, <code>skip_first_frames=0</code>, <code>select_every_nth=1</code>. 
- Extractor: <code>name=Action_01</code>, <code>mode=Full Reference</code>, <code>concept_type=pose_motion</code>, <code>ref_resolution=512</code>, <code>latent_frames=56</code>. 
- Other options: <code>pool_h=16</code>, <code>pool_w=16</code>, <code>motion_only=false</code>, <code>save=true</code>, <code>subfolder=motion</code>. 
 
Suggested workflow: 
 
Upload your own usable action footage and select it in <code>VHS_LoadVideo</code>. Check that the first 121 loaded frames adequately cover the motion you want. The 56 latent frames setting is separate from the raw-video frame cap; do not confuse either with a guaranteed clip duration. Run the extractor and inspect <code>motion/Action_01</code> under <code>models/refmods/</code>. If static appearance from the source video is transferred too strongly, experiment with <code>motion_only</code>—but the supplied JSON leaves it off. Later choose this saved motion asset in the Apply or Text Encode generation path. 
 
Limits 
 
Reference-driven motion cannot ensure precise joint positions, actor-actor contact or a clean separation of the original performer's clothing from the motion. The action-filmstrip cover is artwork explaining the task, not a produced frame sequence or tested result. 
 
Dependencies, tutorial and deployment 
 
This project uses ComfyUI, MiniMax H3 weights and the <a href="https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod" rel="ugc">upstream RefMod node pack</a>; workflows involving <code>VHS_</code> nodes require <a href="https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite" rel="ugc">Video Helper Suite</a>. See <a href="https://huggingface.co/Comfy-Org/MiniMax-H3" rel="ugc">Comfy-Org's H3 model inventory</a> for official distribution guidance. The exact int8-convrot video VAE and Turbo 4-step LoRA named in the creator's JSON need separate availability/provenance checks. We distribute this workflow JSON and its original cover, not external checkpoints or third-party media. Respect applicable rights and licenses. 
 
<a href="https://www.compshare.cn/images/323moGjpvfeM?referral_code=A7fTQUYY9lCDFJwA6MMvoq" rel="ugc">Deploy AI-KSK's H3｜RefMod免训练超级参考 image on 优云智算</a>. Referral disclosure: the deployment link includes an AI-KSK referral code and may support the creator. Check the cloud provider's actual billing and environment before deploying. 
 
<a href="https://youtu.be/H5L5sO7dRu8" rel="ugc">Watch the complete six-workflow YouTube tutorial</a>. This project is one module in that episode. Description and settings were prepared by inspecting the source JSON and transcript, not from a new measured benchmark. 
 
Bilibili tutorial: https://www.bilibili.com/video/BV1F2nZ6SEA7/

Files with SHA256: 5c40bc09512c5f928c17595f6c04e271e3386917c5bea32e56da1cb502a15f90