MiniMax H3 Martial Arts Workflow (Multi-Keyframe)

RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2095907246101291010/?inviteCode=rh-v1111

What this graph is

MiniMax H3 Martial Arts Workflow (Multi-Keyframe) for MiniMax H3 martial-arts motion. 138 nodes, 152 links, 25 groups. Supplied file only - no render, no timing and no quality comparison was produced for this listing.

LoRA stack (saved values)

• Minimax H3真实电影质感V0.1_FeiHouRemix_v0.6_compat_v001_T8.safetensors at 0.5 model / 1.0 clip
• camera_motion_h3_lora_v1_3000_pruned.safetensors at 0.3 model / 1.0 clip
• wushu_spatial_physics_clean_2500.safetensors at 0.3 model / 1.0 clip
• wushu_action_fl2va_v7_2000step_no_adaln.safetensors at 0.4 model / 1.0 clip

These are saved widget values, not tuned recommendations. Test them against your own footage.

Sampler (saved values)

seed 755708092963751 (fixed), 20 steps, CFG 1, euler / simple, denoise 1.

Note the mismatch worth checking: the stack loads the 8-step turbo LoRA at 0.8 while the sampler array says 20 steps. Confirm the effective step count in your own interface.

Files

• Workflow JSON: MiniMax H3 武打工作流（多关键帧）.json
• Cover: MiniMax H3 武打工作流（多关键帧）.jpg (AI-generated promo frame, platform watermark, not a graph output)

Tutorial and rights

Source video (not content-verified here): https://www.youtube.com/watch?v=3Ak4mkrhOrY. Chinese mirror: https://www.bilibili.com/video/BV1rBty6AENs. Weights, input clips and audio are not bundled; check each licence separately.