<h3>🖐️ CROWN Perfect 5-Finger Hands, Eye Catchlights &amp; Human Anatomy Booster | MiniMax H3</h3>
[English] 
Trained by NineCrowns (<code>[CROWN]</code>) on 100% unwatermarked 1080P/4K high-shutter Douyin &amp; TikTok close-up finger-dance (<code>手势舞</code>) and expressive hand-gesture footage at True 768P (<code>768×1344</code>) on an <code>8×NVIDIA B200</code> cluster (<code>Rank 64 / Alpha 64, Video-Only Blocks 10–49</code> with Ostris De-Distillation Adapter). 
Specifically engineered to fix MiniMax H3's hand motion-blur, fused/mutated fingers, and dead/glassy eyes during dancing, vlogs, and close-up acting:
<ol>
 <li>1/1000s High-Shutter 5-Finger Hand Articulation (<code>crown_perfect_hands_eyes</code>): Locks five anatomically distinct, well-separated fingers on each hand with crisp knuckle contours and clean manicured fingernails during fast finger-hearts, V-signs, hand waves, and cheek-cupping poses—eliminating 6-finger mutations and slow-shutter hand smearing.</li>
 <li>Luminous Eye Iris Catchlights &amp; Natural Physiological Blinking: Enhances crystal-clear iris reflections, natural 3-second blinking rhythm, and expressive winks without facial distortion.</li>
 <li>Universal Compatibility: Works standalone or stacked alongside character/style LoRAs across <code>Ref2VA</code>, <code>I2VA</code>, and <code>T2VA</code>.</li>
</ol>
[中文说明] 
由 NineCrowns (<code>[CROWN]</code>) 基于 100% 无水印原生 1080P/4K 抖音原相机高快门怼脸手势舞、美甲手指律动与近景微表情实拍素材，在 <code>8×NVIDIA B200</code> 集群上以 真 768P（<code>768×1344</code>） 分辨率、<code>Video-Only (Blocks 10–49, Rank 64 / Alpha 64)</code> 架构精炼而成。专门攻克 MiniMax H3 (Ref2VA / I2VA / T2VA) 手部模糊畸变与眼神呆滞痛点：
<ol>
 <li>1/1000s 高快门五指解剖结构锁（<code>crown_perfect_hands_eyes</code>）：在双手比心、脸颊比耶、十指张开挥手、捧脸杀等高频手势运动中，强制左右手各 5 根手指根根分明、指节与美甲轮廓锐利清晰，彻底根除多指/融指畸变与慢快门手部残影。</li>
 <li>清澈虹膜眼神光与自然生理眨眼：强化晶状体通透高光反射、每 3 秒自然轻柔眨眼与俏皮单眼 Wink，杜绝眼神空洞死板。</li>
 <li>全模式通用增强：支持在 <code>Ref2VA</code>、<code>I2VA</code> 与 <code>T2VA</code> 中单挂或与其它风格/动作 LoRA 叠加使用。</li>
</ol>
<hr />
📊 Real Production Test Configuration / 真实测试参数配置
<ul>
 <li>Base Model (底模): <code>minimax_h3_ref2va_bf16.safetensors</code> + <code>qwen3vl_32b_minimax_h3_bf16.safetensors</code> + <code>minimax_h3_video_vae_fp32.safetensors</code></li>
 <li>Resolution &amp; Sampler (分辨率与采样器): <code>768×1344</code> (or <code>1344×768</code>), <code>25 steps</code>, <code>euler</code> + <code>beta</code>, <code>cfg: 1.0</code>, <code>MiniMaxH3SigmaShift (shift_video: 12.0, shift_audio: 3.0)</code></li>
 <li>LoRA Weight (单挂本 LoRA 推荐权重): <code>0.70 ~ 0.80</code> (Showcase video generated with ONLY this LoRA at <code>0.75</code>, zero third-party LoRAs)</li>
 <li>Trigger Word (触发词): <code>crown_perfect_hands_eyes</code></li>
</ul>

CROWN Perfect 5-Finger Hands, Eye Catchlights & Human Anatomy Booster | MiniMax H3

[EN] v1.0 True-768P (768×1344) Video-Only Rank 64 (Blocks 10–49) release. Recommended weight: 0.70–0.80. Locks 1/1000s high-shutter 5-finger hand anatomy, crisp knuckles/fingernails, and luminous eye catchlights. [中文] v1.0 真 768P (768×1344) Video-Only Rank 64 (Blocks 10–49) 正式版。推荐权重：0.70–0.80。强力锁死 1/1000s 高快门五指解剖结构、清晰指节美甲与清澈虹膜眼神光。

Files with SHA256: 68f425f6bb1e38091f4582504edf84584e3ea24a1e53e082c8a540c8ecd2f07a