⚔️ CROWN 3D Spatial Crossing & Five-Finger Hilt Grip Master | MiniMax H3

[English]
Trained by NineCrowns ([CROWN]) on 100% real-world human martial-arts & dramatic blocking footage using an 8×NVIDIA B200 cluster (Rank 64 / Alpha 64, Video-Only Blocks 10–49 with Ostris De-Distillation Adapter).
This LoRA solves two of the hardest structural challenges in MiniMax H3 (Ref2VA / I2VA / T2VA):

1. Five-Finger Bottom-Hilt Grip & Rigid Unbending Blade Physics (crown_grip): Locks all five fingers cleanly around the weapon handle below the crossguard (preventing characters from grabbing the sharp blade edge or floating the hilt) and maintains ruler-straight, unyielding forged-steel blade geometry during linear cuts, thrusts, and edge-to-edge parries.
2. Multi-Character 3D Foreground-Background Spatial Crossing Without Identity Swap (crown_spatial): Enables two characters in Ref2VA to charge past each other in an X-shaped foreground-background occlusion swap (Left↔Right position exchange) while 100% preserving each character's individual facial bone structure and outfit without face-swapping or clothes-bleeding.

[中文说明]
由 NineCrowns ([CROWN]) 基于 100% 纯真人实拍武术兵击与双人走位素材，在 8×NVIDIA B200 集群上采用 Video-Only (Blocks 10–49, Rank 64 / Alpha 64) 架构与反蒸馏 Adapter 训练而成。专门解决 MiniMax H3 (Ref2VA / I2VA / T2VA) 的两大核心痛点：

1. 五指标准正握剑柄与刚性笔直剑身锁（crown_grip）：强制五指完整包握护手下方的剑柄（根除手抓剑刃、剑悬浮脱手），并在直线斜劈、刺击与十字格挡中保持刚性冷钢剑身笔直不发软。
2. 双人三维前景-后景 X 交错换位防串脸（crown_spatial）：支持双角色在 Ref2VA 模式下完成前景遮挡横向交错换位（如左男右女交错后换为左女右男），全程 100% 锁死各自五官与服装，0 串脸、0 换装。

• Base Model (底模): minimax_h3_ref2va_bf16.safetensors + qwen3vl_32b_minimax_h3_bf16.safetensors + minimax_h3_video_vae_fp32.safetensors
• Sampler & Steps (采样器与步数): 25 steps, euler + beta, cfg: 1.0, MiniMaxH3SigmaShift (shift_video: 12.0, shift_audio: 3.0)
• LoRA Weight (单挂本 LoRA 推荐权重): 0.65 ~ 0.75 (Showcase video generated with ONLY this LoRA at 0.65, zero third-party LoRAs)
• Trigger Words (触发词): crown_spatial_grip, crown_grip, crown_spatial