15-Second Action Scene Timeline and Continuity Prompt Tips for Pollo AI

AI video creator Liora Solveil (@LioraSolveil) has shared a battle-tested prompt framework for generating realistic, continuous 15-second martial arts action scenes with Pollo AI (@polloai_kr). Inspired by action choreography from gloom.creations and recreated by Liora, this workflow locks the lead character's identity and outfit using a single reference image while directing close-quarters fight choreography, strict weapon constraints, and progressive environmental destruction across an explicit second-by-second timeline.

Image source: @LioraSolveil via X

When directing complex multi-character (1 vs. 5) melee combat in generative video models, creators frequently encounter severe visual anomalies: limb merging, unauthorized weapon multiplication, supernatural teleportation, and background resets across continuous takes. Liora's prompt provides a systematic architecture combining character image locking, hard constraints on weapons and physical space, and precise second-by-second action beats.

1. Character, Weapon, and Spatial Hard Locks

The upper section of the prompt enforces strict boundaries to prevent the model from hallucinating or randomizing elements mid-take:

• Lead Character Reference Lock: Directs the engine to treat Image 1 exclusively as the identity and outfit anchor—preserving the heroine's pink bob hair, facial features, physical proportions, and minimalist white athletic outfit. Crucially, it instructs the model not to replicate any static pose from the reference image, allowing fluid movement.
• Text-Only Enemies: The five opponents are specified solely via text without reference images. Defining them as yakuza enforcers from the same syndicate wearing unified black dojo uniforms with visible irezumi tattoos gives the engine a coherent group identity without overloading image reference slots.
• Weapon Lock: To prevent weapons from spontaneously multiplying or switching types, the prompt strictly assigns the katana exclusively to enemy A1, while enemies A2 through A5 each carry exactly one hand axe. An explicit negative constraint prohibits giving additional swords or randomizing weapons.
• Single Continuous Environment (Space Lock): The entire sequence is locked into a single traditional Japanese wooden hall with shoji screens, a decorative folding screen, low tables, and a glass cabinet. Location jumps and background resets are forbidden.

2. Complete 15-Second Action Prompt

Below is the complete verbatim English prompt shared by Liora. It can be copied directly into compatible video generation engines:

REFERENCE

Use Image 1 as the identity and outfit lock for the female lead only. Preserve her pink bob hair, face, body proportions, white sporty outfit, and overall recognizable identity throughout the entire video. Do not copy any pose from the image.

No reference image is used for the enemies; define them from text only.

SUBJECT

A fast photorealistic martial arts action scene inside a Japanese traditional wooden interior.

The main character is a pink-haired young woman, agile, calm, lethal, and highly skilled in close combat. She wears the white outfit from Image 1 and remains visually consistent for the entire video.

Her opponents are five male yakuza enforcers, all from the same organization, all wearing black dojo-style uniforms. Their bodies are covered in Japanese irezumi tattoos visible at the neck, chest opening, forearms, wrists, and hands. They are larger and heavier than the heroine and visually unified as one gang.

WEAPON LOCK

Enemy A1 carries the only katana in the entire scene.

Enemies A2, A3, A4, and A5 each carry one hand axe.

Do not randomize weapons. Do not give extra swords to anyone else.

Maintain weapon continuity throughout the full scene.

BACKGROUND / SPACE LOCK

One single continuous fight space only.

A warm-lit Japanese wooden hall with polished wooden floor, shoji-style walls, structural wooden columns, a decorative folding screen, a low wooden table, ceramic decor, and a glass display cabinet.

This is one continuous physical space. No location jumps. No background resets.

TIMELINE

0.0–1.7s

The pink-haired heroine stands in the room. A1 attacks first with a powerful overhead katana strike from behind or a rear diagonal angle. She ducks at the last possible moment and slips out of the blade path, then immediately moves into his flank.

1.7–2.7s

She controls A1 at the neck and shoulder line, rotates her torso, and throws him hard to the floor. The takedown is fast, grounded, and physical. She uses the momentum of the fall to transition immediately into the next action.

2.7–4.0s

A2 rushes in with a hand axe. She closes distance and launches a fast jumping spinning kick. Her foot clearly connects with A2’s upper body. A2 flies backward into the decorative folding screen. The screen collapses and breaks apart.

4.0–5.6s

A3 comes in at close range with a hand axe. She evades with a short step and upper-body slip, then counters with compact close-range strikes: a body shot, a sharp elbow-like hit, and a fast redirect that breaks his pressure and knocks him off balance. He staggers away but is not yet fully finished.

5.6–7.9s

A4 charges with a hand axe. She slips low under the attack, catches or redirects the attacking arm, controls the wrist and forearm, rotates the body line, and drives him hard sideways into the glass display cabinet. The cabinet shatters violently, with visible glass explosion and debris.

7.9–10.0s

A5 enters as the last major attacker with a hand axe. At close range she lands a rapid combination, then finishes with a strong spinning kick or turning kick that launches A5 backward across the room.

10.0–11.4s

A5 crashes down near the furniture. The heroine regains her stance immediately. A3, still active, suddenly re-enters from the rear diagonal with his hand axe for one final desperate attack.

11.4–12.8s

She evades A3’s axe line with a tight body movement, controls the wrist, and ends him quickly with a short punch, a body shot, one knee or elbow, and a final shove kick or side kick that throws him backward into the low table area. He falls and is fully defeated.

12.8–15.0s

The fight is over. The heroine stands upright and slowly settles her breathing. The camera gives a subtle push-in or a short gentle arc around her. Around the room lie the defeated yakuza, dropped hand axes, A1’s katana, broken ceramic fragments, the collapsed folding screen, shattered glass, and scattered debris. She finishes in a calm, cold, controlled hero shot, looking forward or slightly off-axis.

ACTION STYLE

Fast real-contact cinematic fight choreography.

No superpowers. No fantasy aura. No teleportation.

Every hit must read clearly: attack path, evasion, contact, force transfer, reaction, and follow-through.

Use grounded martial arts physics with strong momentum continuity.

She must never pause unnecessarily between attacks.

All movements should feel efficient, sharp, and lethal.

CAMERA

A dynamic handheld cinematic camera that feels like one continuous action shot.

Use close-range tracking, short whip pans, brief push-ins, and natural reframing to follow the action.

The camera is an observer. Maintain spatial clarity.

Do not flip left and right orientation.

Do not use excessive orbiting or slow motion.

The room layout must remain readable throughout.

STYLE

Photorealistic live-action action cinema.

High detail skin, hair, cloth, tattoos, broken wood, glass shards, and weapon surfaces.

Warm cinematic interior lighting with strong contrast between the heroine’s white outfit and the gang’s black uniforms.

The pink hair must remain a strong visual focal point.

AUDIO

Crisp cloth movement, foot friction, heavy body impacts, blade swishes, axe swings, wooden collisions, folding screen break, glass shatter, and brief breathing sounds.

No dialogue.

The sound should emphasize contact timing and environmental destruction.

CONSTRAINTS / HARD LOCKS

Female lead identity must remain consistent with Image 1 at all times.

All five enemies are male yakuza with black dojo-style uniforms and visible Japanese tattoos.

A1 is the only enemy with a katana.

A2, A3, A4, and A5 each carry one hand axe only.

Do not add more enemies.

Do not remove the destroyed environment continuity.

Keep everything in one continuous Japanese wooden hall.

No slow dramatic posing during the main fight.

The final 2 seconds must be a payoff ending, not a new fight wave.

End on a strong hero shot with the aftermath visible.

3. Directing Grounded Physics and Camera Movement

The directive sections at the bottom of the prompt instruct the engine on tactile cinematic realism:

• Realistic Combat Physics (Action Style): Prohibits supernatural auras, energy blasts, or teleportation. Forces grounded momentum transfer where each strike exhibits clear initiation, point of impact, target reaction, and recovery.
• Handheld Continuity (Camera): Mandates dynamic tracking and subtle whip pans that follow combatants without crossing the 180-degree axis line. Unnecessary slow motion and orbiting camera turns are explicitly blocked to keep spatial relationships intuitive.
• Cumulative Environmental Aftermath: Debris from broken props—such as the folding screen collapsed during 2.7–4.0s and the glass display cabinet shattered during 5.6–7.9s—must remain visible across the room through the final hero shot at 12.8–15.0s.

4. Practical Caveats and Limitations

Before deploying this framework into a production pipeline, consider the following constraints:

1. Complex Melee Artifacts: Dense multi-character fighting and shattering glass require substantial generative capability. Depending on the model weights and resolution, momentary limb fusion or object morphing artifacts may occur.
2. Character Reference Capability Required: Consistent depiction of the heroine's face, hair color, and outfit requires an image-to-video or reference-locking feature (such as Pollo AI's character reference engine).
3. Temporal Prompt Awareness: Sub-second timeline instructions (e.g., 0.0–1.7s, 1.7–2.7s) yield the highest fidelity in state-of-the-art video architectures designed to interpret chronological scene beats.

Original source

• Liora Solveil (@LioraSolveil) on X: https://x.com/LioraSolveil/status/2110719565793730847