Introduction¶
The common failure in using video generation models for fight scenes isn’t the absence of visuals, but the absence of actual combat within the visuals. Characters pose individually, taking turns to display their moves, resulting in a “standing dummy” appearance. The root cause is that the prompt translates “there are three people” into “three people appear in succession,” giving the model instructions for a sequential showcase rather than a chase and suppression sequence.
The Skill introduced below compiles a one or two-sentence Chinese requirement into a set of prompts describing movement, pursuit, contact, guard breaking, and finishing moves. It targets genres like Xianxia, Fantasy, and high-octane ultimate moves. It delivers pure text and does not call any APIs.
What This Is¶
Silver Fight Video Prompt Director, located at @user_c5db8462/silver-fight-video-skill, published and maintained by community developer user_c5db8462. Current version 1.0.4, categorized under content-creation, and tagged with community source.
It solves the problem where users only have a simple sentence like “Xianxia fight, protagonist fights three, uses an ultimate at the end,” but the videos generated by general prompts lack physical interaction. This Skill breaks down the requirement into executable cinematographic conditions and outputs them as prompt text.
Core Capability: Physical Conditions Mandated in Prompts¶
Silver’s prompts do not describe emotions; they describe events along a timeline. The following are the hard conditions it writes into the prompts.
Pressure Relay, Not a Queue for the Slaughter¶
The three enemies do not follow a “finish the first one before the second attacks” pattern. The prompt requires: after the first enemy attacks and slides back or loses balance, the second enemy steps into the gap to continue pressure; once the second is deflected, the third enters. Throughout the segment, one enemy remains in the attack position at all times, while the other two compress the protagonist’s movement space.
At Least Three “Dual Threat” Frames in the First 8 Seconds¶
Within the first 8 seconds of the video, arrange at least three compositions where two threats appear in the frame simultaneously. The rule is: the current engagement hasn’t ended, but the next threat has already entered the shot, rather than waiting for the current round to finish before cutting.
Ultimate Move Must Be Triggered in Motion¶
Ultimate moves cannot be initiated from a stationary position. The prompt requires the ultimate to leave the caster, cross the battlefield, encounter countermeasures en route, and finally alter the enemy’s final state (knockdown, bounce, disintegration, etc., depending on the genre). A static frame with light effects does not satisfy this.
Strike Effects Layered by Intensity¶
Effects are not piled up until the very end. The prompt delivers effects layered by strike intensity: 3 layers for normal hits, 4 for heavy hits, and 5 for ultimates. The effects for each layer appear during the timeframe of the corresponding strike, rather than erupting all at once at the end of the clip.
Three Continuous Hits in the Last 3.5 Seconds¶
Arrange three continuous hits on the target within the final 3.5 seconds of the ending segment. Each enemy must have a visible impact point lasting over 0.35 seconds—the position of being knocked back, the destination of being knocked flying, or the pose of falling to the ground. The prompt explicitly excludes writing that substitutes water mist, glare, or camera shake for final states, because while such effects look impactful, they do not change the enemy’s final state.
Six Built-in Routings¶
The Skill has six built-in routings; select one based on the requirement:
- High-speed Martial Combat: Fast-paced, dense contact regular battles.
- Xianxia Amplification: Xianxia battles with sword riding, magic, and artifacts.
- Standalone Ultimate: Skips the preceding combat and goes straight to the ultimate.
- Combat to Ultimate: Progressive buildup in the first half, ending with an ultimate.
- Continuous Clips: Outputs multiple continuous combat segments.
- Failure Repair: Diagnoses issues in existing videos and provides repair prompts.
Element Ultimate Library & Weapon/Movement Library¶
Includes a built-in library of 13 elemental ultimates and weapon/movement styles to provide specific material when writing ultimates and movement, avoiding generic descriptions like “sword energy” or “flash” in every output. The significance of elemental classification is to allow for counter and combination relationships between ultimates, enabling the writing of specific reactions when countered.
SIGNATURE_FULL Capability Mode¶
The Skill provides a SIGNATURE_FULL tier, enabled by default, outputting with full capability without compromise. This means that by default, all the physical conditions, routings, element libraries, and movement libraries described above are active, without requiring the user to explicitly declare “full mode.”
Diagnosing and Repairing Existing Videos¶
Besides generating from scratch, the Skill also supports diagnosing failed existing videos, covering these common problems:
- Standing Dummy Performance: Characters take turns showcasing moves, lacking displacement and pressure.
- Slow Motion: The overall rhythm is sluggish, and contact is drawn out.
- Fake Hits: The moment of impact only has effects, with no change in the enemy’s posture.
- Sticker-style Effects: Light effects are pasted on the character’s surface and do not move with the action.
- Ultimate Obstruction: The ultimate is cut away from by the camera, blocked by foreground objects, or compressed into a fleeting flash at the end.
The repair path corresponds to the Failure Repair routing mentioned in the six routings above.
Delivery Method¶
The output is pure text prompts, calling no APIs. The Skill page tag is requires_api_key=false, meaning no API Key is needed. The invocation of the video model itself (e.g., Seedance) is not within the scope of the Skill; the Skill is only responsible for compiling requirements into prompt text.
Installation and Activation¶
The official page retrieved did not return installation commands, directory structures, or configuration fields, so no speculation is included here. For installation, please refer to the instructions on the official page https://www.skillhub.cn/skills/user_c5db8462/silver-fight-video-skill.
Use Cases and Notes¶
Suitable for these requirements: Xianxia fights, fantasy battles, 1 vs. many, high-octane action, impact feel, ultimate effects, weapon arts, elemental tactics, combat storyboard prompts, and fight video repair. The trigger words include Seedance, indicating the author expects the output to be used for this type of video model.
Two points to note during use. First, the Skill outputs prompt text; the final video quality still depends on the video model’s own ability to restore displacement, contact, and final states. If the model cannot create dual-threat frames or layered effects, the prompt cannot conjure them out of thin air. Second, the physical conditions in the prompt are written according to a timeline (first 8 seconds, last 3.5 seconds, 0.35-second impact points). If the target video duration is very short, these conditions will squeeze each other, requiring adjustment based on the actual duration.
Conclusion¶
The value of Silver Fight Video Prompt Director lies in translating the phrase “make the fight look good” into executable cinematographic conditions, and defaulting to full capability output. It does not generate videos or call APIs; the deliverable is a prompt ready to be pasted into a video model.
Official Address: https://api.skillhub.cn/user_c5db8462/silver-fight-video-skill
Page Address: https://www.skillhub.cn/skills/user_c5db8462/silver-fight-video-skill
