Reproducción automática
Ejecutar en local
Motion_Repair_V2.safetensors
148 MB · SafeTensor · bf16
Las descargas las sirve Civitai. Algunos creadores piden iniciar sesión en Civitai primero.
Sobre este modelo
MINIMAX H3 · LoRAs
by FourBunny - This is a general-purpose motion support LoRA, not a combat-only LoRA. It can be used for running, sports , dance, acrobatics, character interaction, combat, weapon motion, and other dynamic scenes.
Notas del creador
General Motion Continuity Repair LoRA
🎬 BUNNY Motion Repair V2 — High-Difficulty Motion Shouldn't Be a Lottery.
This is a general-purpose motion support LoRA, not a combat-only LoRA. It can be used for running / sports / dance / acrobatics / character interaction / combat / weapon motion and other dynamic scenes.
这是一个通用动态辅助 LoRA,并不是战斗专用模型,可用于跑步 / 体育 / 舞蹈 / 翻滚跳跃 / 人物互动 / 战斗 / 武器动作等动态场景。
🚀 Run It on RunningHub
🎁 New users on RunningHub International can get 1000 RH Coins after registration.
🔥 What’s New in V2
-
Targeting Failure Zones: Focuses on MiniMax H3's hardest movement areas—rapid flips, spins, rolls, inversions, and extreme orientation recovery.
-
Rotational & Inverted Upgrade: Significantly expands full-body rotational training while keeping V1’s general motion repair power.
-
Smooth Full Sequences: Enhances the complete chain: Takeoff ➔ Inversion ➔ Continuous Spins ➔ Reorientation ➔ Landing, not just the middle frames.
🤸♂️ Noticeable Motion Gains
-
Power Moves & Ground Spins: Thomas flares, windmills, airflares, and ground rotations stay stable without collapsing during repeated cycles.
-
Airborne Continuity: Flips, twisting jumps, and aerial spins maintain structural integrity—no more twisted body masses, misconnected limbs, or out-of-sync head-torso rotation.
💡 Under the Hood
-
Capability-Oriented Training: Learns structural stability, orientation awareness, and momentum recovery rather than memorizing trick names, ensuring strong generalization to unseen motions.
-
Usable Output First: As a repair LoRA, rare imperfect frames can still happen in extreme transitions, but it reliably shifts generations from discarded clips to usable shots.
💬 In Short
High difficulty, higher stability. Flips stay coherent, spins stay together, and landings stop falling apart. (高难动作,稳定性跟上。空翻不扭、旋转不散、落地不崩!)
Recommended Usage / 推荐使用
Standalone use: around 0.9 is currently recommended.
单独使用时,目前推荐权重约为 0.9。
When used alone, Motion Continuity Repair needs a relatively high weight to produce a clear effect on motion continuity / slow-motion tendency / missing transitions / prompt-following.
单独使用时,需要相对较高的权重,才能比较明显地影响动作连续性 / 慢动作倾向 / 动作断链 / 提示词遵循度。
0.5–0.70.5–0.7
Because Combat already provides strong motion speed and impact, Motion Continuity Repair only needs a medium weight to correct continuity / coordination / action logic without unnecessarily taking over the motion.
因为 Combat 已经提供了较强的动作速度和打击感,此时 Motion Continuity Repair 只需要中等权重来修正动作衔接 / 协调性 / 动作逻辑,避免过度接管整个动态。
More complex sequences can gradually increase the Stage 1 weight, but very high weights may begin to reshape the entire motion logic / affect visual style / or alter audio characteristics.
动作越复杂可以逐渐提高一采权重,但过高时可能开始重构整段动作逻辑 / 改变画面表现 / 影响声音特征。
Stage 2: around 0.2–0.3 is recommended.
二采建议约 0.2–0.3。
Higher Stage 2 weights may over-stabilize the sequence and reduce motion intensity / speed / impact.
二采权重过高可能会把动作收得太稳,从而削弱动作幅度 / 速度感 / 冲击力。
最简单地说就是:
单独使用:≈0.9 / 搭配 Combat:0.5–0.7 / 二采:0.2–0.3
这个才是目前测试出来的正确推荐方式。
Although it is a general-purpose LoRA, it works especially well together with a Combat LoRA. Combat provides speed and impact / Motion Continuity Repair focuses on continuity, coordination, prompt-following and interaction logic.
虽然它是一个通用 LoRA,但目前非常推荐与 Combat LoRA 搭配使用。Combat 负责速度与打击感 / Motion Continuity Repair 负责动作衔接、协调性、提示词遵循与交互逻辑。
Trigger Word / 触发词
Trigger word: bunny_crisp_motion — optional. It mainly serves as an explicit identifier for the LoRA and is not required for normal use.
触发词:bunny_crisp_motion —— 可选。它主要用于明确标识这个 LoRA,正常使用时并不是必须填写。
You can test with or without the trigger word depending on your workflow and prompt.
你可以根据自己的工作流和 Prompt,自行测试是否加入触发词。
What does it do? / 它主要做什么?
Ever had a great clip ruined by less than one second of strange motion? A sudden slowdown / a missing action / an awkward limb trajectory / a broken interaction can easily make an otherwise usable clip unusable.
你是否遇到过这种情况:一段视频明明整体很好,却偏偏有不到 1 秒突然慢下来 / 漏掉一个动作 / 肢体轨迹变得诡异 / 交互突然断掉,最后整段素材因此报废?
This LoRA is designed around repairing those moments rather than simply making every motion stronger.
这个 LoRA 的重点就是修复这些“突然掉链子”的瞬间,而不是粗暴地把所有动作统一增强。
Current A/B testing shows improvements in unwanted slow-motion behavior / missing motion transitions / abnormal limb trajectories / action coordination / prompt-following / interaction continuity / and some mismatched motion-feedback events.
目前 A/B 测试中已经观察到对 异常慢动作 / 动作断链 / 诡异肢体轨迹 / 动作协调性 / 提示词遵循度 / 交互连续性 / 部分错误动作反馈 的改善。
When H3 already handles a scene well, the difference may be very small / when the base model begins to struggle, the difference can become much more obvious.
当 H3 本来就很擅长某段动态时,开启前后的区别可能非常小 / 当底模开始掉速、断动作或失去逻辑时,差异反而会明显很多。
It doesn't simply make H3's strengths stronger — it tries to keep its weaker motion states from falling apart.
它不是单纯让 H3 的强项更强,而是尽量不让 H3 在弱项动态里突然掉下去。
Current A/B Results / 当前对比表现
Different weights can progressively change motion intensity and eventually begin reshaping the overall motion rhythm.
随着一采权重提高,可以逐步改变原本偏弱的动态表现,高权重下甚至会开始重构整体动作节奏。
In complex action sequences, the LoRA can reconnect missing actions and reduce strange limb paths, making the motion chain more complete.
在复杂连续动作中,它可以补回部分没有连接上的动作,并减少诡异肢体轨迹,让动作链更加完整。
In scenes where even a Combat LoRA cannot fully suppress H3's slow-motion tendency, Motion Continuity Repair can further pull the sequence back toward a faster real-time motion state.
在一些连 Combat LoRA 都无法完全压住慢动作倾向的场景中,Motion Continuity Repair 还能进一步把动作往正常高速节奏拉回。
It can also improve action-prompt adherence, turning sequences that originally move randomly into results that follow the intended action order more closely.
它也可以提升动作提示词遵循度,让原本只是胡乱运动的结果,更容易按照 Prompt 中指定的动作顺序执行。
Weapon Interaction / 武器交互
A portion of weapon finisher / execution animation data was included in the dataset.
训练数据中还加入了一部分武器处决 / 终结动画。
Current testing shows some improvement in weapon disappearance / clipping / unstable fast swings / incomplete hit reactions / and certain finishing o
SHA256
EDFA0E2858CAEBA7BE9108883F61DA8525D3C58AF998C94AC7A449D9603D7574
ID de modelo de ModelTrends.ai
#26824
Denunciar contenido
Renders de ejemplo
Prompts y ajustes compartidos por quienes hicieron estos renders en Civitai. Elige un render para ver cómo se hizo.
Ajustes
Pasos
8
Guidance
1
Tamaño
1584x1200
Prompt
integrated_multimodal_description: [Shot 1] 00:00.000–00:03.000. Live-action, gritty dark fantasy, overcast grey daylight in a foggy cypress swamp draped with Spanish moss. Starting from the opening frame: the bearded knight in dented, rust-streaked plate armor, an open-faced steel helmet, and a torn green tabard stands knee-deep in murky water, swinging his longsword with a round shield on his arm, as a three-headed hydra with grey-green scales and spiked crests rears up out of the water in front of him, all three jaws open and dripping. A rickety wooden dock with a lit hanging lantern stands to the right. The camera tracks in fast at water level. At 00:00.600 his blade slams into the left head's neck, spraying water and the head recoils with a shriek. At 00:01.800 the center head strikes straight down at him; he throws up his shield and the jaws clamp onto its rim, wrenching him sideways in a burst of spray.
[Shot 2] 00:03.000–00:05.500. Hard cut to a low shot skimming the water surface. The right head snakes low and fast across the swamp, jaws snapping at the small green frog; at 00:03.800 the frog leaps onto the rusted helmet half-sunk in the water, and at 00:04.600 springs again onto the dock planks just as the jaws crunch shut on empty air, sending the half-sunk helmet spinning away.
[Shot 3] 00:05.500–00:08.000. Hard cut to a medium shot on the knight. He rips his shield free of the center head's jaws, spins, and at 00:06.200 smashes the shield edge into its snout, knocking the head aside. The left head lunges back in; he ducks under it, water churning around his armored legs. The knight with a rough, breathless voice (S1) snarls: <d>[English] Come on, then, you ugly beast!</d>
[Shot 4] 00:08.000–00:10.000. Hard cut to a wide shot of the swamp. All three heads rear up together, roaring. The knight charges forward through the water, sword raised high, kicking up spray with every stride. At 00:09.200 the right head whips sideways and smashes through a dock post, the lantern swinging wildly and wooden splinters flying, as the knight closes in on the hydra's body.
overall_soundscape: Sloshing water, dripping moss, and croaking swamp insects throughout. A heavy steel chop and a piercing reptilian shriek at 00:00.600, a crunching clang of teeth on the shield at 00:01.800, the frog's croak and small splashes, a hollow metallic ping as the helmet is knocked, a wet snap of empty jaws, a dull metallic smash of shield on snout at 00:06.200, the knight's heavy breathing, triple overlapping roars, splashing strides, and a splintering crack of timber with the lantern's chain rattling at 00:09.200.
non_diegetic_music: A gritty, fast-tempo orchestral score with low pounding drums, harsh brass, and tense strings, hitting accents on each strike and building to the final charge.
Modelos similares
Modelos de IA para generar imagen, vídeo y texto
motion repair h3 es un LoRA para la familia MiniMax H3 listado en ModelTrends.ai, un catálogo de solo lectura de modelos de IA de imagen, vídeo y texto de código abierto. Compáralo con otros modelos MiniMax H3, consulta su puntuación de calor para ver su tendencia y ábrelo en PirateDiffusion o BitVector para probarlo.
Explorar por familia de modelos
Modelos Text / LLM
411
Modelos Anima
509
Modelos Chroma
14
Modelos Flux
1,071
Modelos Flux 2 / Klein
1,237
Modelos MiniMax H3
331
Modelos Hunyuan
197
Modelos Ideogram
20
Modelos Krea2
674
Modelos Qwen
17
Modelos Qwen2
36
Modelos Ltx2
150
Modelos Stable Diffusion 1.5
6,465
Modelos Stable Diffusion XL 1.0
15,432
Modelos Zimage
804
Modelos Wan
919

Model
Trends
.ai
© ModelTrends.ai
|
Hecho en Japón
|
© 2026
