MiniMax H3 RefMod Motion Reference

https://youtu.be/H5L5sO7dRu8

Creator: AI-KSK · Collection: Video 327 / six-workflow MiniMax H3 RefMod episode

Workflow file: 02_动作RefMod_单视频与文件夹批量.json · Cover file: 武术动作参考-2.png

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 VHS_LoadVideo → the refs_video.ref_video_0 socket on MiniMaxH3RefModExtract → MiniMaxH3RefModInspect → 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: 203719-922186723_small.mp4 — not distributed.
- Video decoder: force_rate=0, custom_width=0, custom_height=0, frame_load_cap=121, skip_first_frames=0, select_every_nth=1.
- Extractor: name=Action_01, mode=Full Reference, concept_type=pose_motion, ref_resolution=512, latent_frames=56.
- Other options: pool_h=16, pool_w=16, motion_only=false, save=true, subfolder=motion.

Suggested workflow:

Upload your own usable action footage and select it in VHS_LoadVideo. 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 motion/Action_01 under models/refmods/. If static appearance from the source video is transferred too strongly, experiment with motion_only—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 upstream RefMod node pack; workflows involving VHS_ nodes require Video Helper Suite. See Comfy-Org's H3 model inventory 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.

Deploy AI-KSK's H3｜RefMod免训练超级参考 image on 优云智算. 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.

Watch the complete six-workflow YouTube tutorial. 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/