AI Fight Generator: Turn an Image Into a Fight Scene

Flick's AI fight generator turns one image, a portrait, a photo or a character sheet, into a fight scene: you set the fighters in a still, write the action, and a video model plays it out.

An ink drawing of a pair of boxing gloves hanging from their laces

Most AI fight generators take a photo and return one random brawl, so the fighters change faces, the punches miss and nothing cuts together into a scene. We read the canvas of Boxing at the Opera, a 17 second short by Hannah L. of the Flick team, where all 18 clips started as a still image and became a boxing match in Seedance 2.5. This page shows those images next to the fight clips they became, the prompts that moved them, and four steps to turn your own image into a fight scene in Flick.

Describe your fight in Flick

Describe the fighters and the action in the box, or pick one of the three examples below, and it opens a new Flick project where you add your image as the first frame.

What does Flick's AI fight generator do?

It takes one image as the first frame of a video and animates the fight you describe in the prompt, from a single punch to a full brawl with a cheering crowd.
  • It animates any still you give it: a photo you upload, a portrait you generate in Flick, or one frame of a scene.
  • It keeps the fighters' faces, costumes and setting from that image, because the clip starts on the image itself.
  • It lets you make each beat of the fight as its own image first, so a punch and the hit it lands are framed exactly as you want them.
  • It holds the whole fight on one canvas, images, clips and cuts together, so a scene grows shot by shot.

Counted on the Boxing at the Opera canvas:

  • 18 video clips, all made with Seedance 2.5 from a still image
  • 17 of those 18 start images were made in Flick with GPT Image 2
  • 17 of the 18 prompts ask for analog film grain, so the shots match
  • 9 of the 18 prompts say "camera locked", so the frame holds still while the fighters move

What does an image look like once it becomes a fight scene?

The image becomes the first frame, and the clip carries on from it: two still figures start trading blows, a stare-down starts to breathe, and a punch lands on the next fighter's jaw.
  • Film: The Miracle by Sam Wasserman, from the Summer 2026 Filmmaker Residency, where one overhead still of two fighters on a rain-soaked tower became a 30 second fight.
  • The whole prompt is three sentences, because the image already holds the fighters, the arena and the camera height.
Overhead view of two small figures facing off on a wet circular tower top ringed with lights at night
Two fighters battle across a rain-soaked circular tower top as the camera orbits around them00:30
One still becomes a whole fightSam Wasserman, director of The Miracle · Seedance 2.5 · Image to video

Seedance 2.5 · the prompt for the clip above

Epic Fight battle between these two. The camera should orbit 360 continously the whole time from this height. Make sure the fight is really intense and action packed. 
  • Film: Boxing at the Opera by Hannah L., a 1940s boxing match staged in a gilded opera house, with every shot animated from a still.
  • Each pair below shows the image on the left and the 5 second clip it became on the right; turn the sound on to hear it.
Portrait of a stocky 1940s boxer with his gloves up in a dark opera hall, his shadow on a red curtain
The same stocky boxer glares at his opponent with his gloves up as the camera slowly moves in00:05
A portrait becomes the stare-downHannah L., director of Boxing at the Opera · GPT Image 2 · Seedance 2.5 · Image to video
Close-up of a sweating boxer with a cut lip in front of stage lights and a red curtain
A close-up of the sweating boxer as he pulls back for a big punch, in slow motion00:05
A close-up becomes the wind-upHannah L., director of Boxing at the Opera · GPT Image 2 · Seedance 2.5 · Image to video
Front close-up of a boxer at the moment a glove hits his cheek, his eyes squeezed shut
A boxer's face twists in slow motion as a leather glove lands on his cheek00:05
A frozen moment becomes the punch landingHannah L., director of Boxing at the Opera · GPT Image 2 · Seedance 2.5 · Image to video
More detail

The start images were made from one another on the canvas: Hannah asked GPT Image 2 for a close-up of one boxer, then for "front facing close up of the guy on the right. same exact moment in time", so the hit is the same moment seen from his side.

Can a character sheet become a fight scene?

Yes: give the video model a sheet of the two fighters as a reference instead of a first frame, and write who throws the first punch, how they grapple and who watches.
  • Xenogenesis, a 17:53 sci-fi horror short by Gertywood from the Summer 2026 Filmmaker Residency, cites one sheet of two miners in 3 of its 420 clips, each a street fight glimpsed from a passing car.
  • The prompt gives the camera its own path, the fight its own beats, and the crowd its own reaction, so all three read in one 5 second shot.
Two bearded miners in yellow rain gear stand face to face in a street at night, about to fight, as a crowd watches00:05
Gertywood, director of Xenogenesis · Seedance 2.0 · Reference to video
Use Miners Who Fight as the visual reference for the two arguing miners.
use Miners Womenand Miners Menas the crowd.
Use Drainage Neighborhood as the visual reference for the ghetto pit-city street and the amber-lit buildings.
Single continuous shot.
Action: close-up on a long lens from the moving car's POV, the camera dollying quickly frame-left to frame-right at the speed of the passing car while panning left to hold the action, catching a street scene in a fleeting passing glance. Two miners face each other in profile, one getting up in the other's face, arguing heatedly, chests nearly touching, on the verge of throwing punches. Then the left miner shoves and swings hits the other miner in the eye. Then they grab each other, grappling in a messy, real, unchoreographed tussle, staggering and wrenching at each other's clothes, no clean martial choreography, just two furious men wrestling on the street. Around them, other miners gather and cheer them on, throwing up their arms and shouting. The camera's lateral speed carries the whole scene through frame quickly, the long lens compressing the action as the car sweeps past and leaves it behind.
Blocking: The two miners face each other in profile at center as the car passes laterally; the camera dollies right at car-speed while counter-panning left to hold them in frame just long enough to catch the shove and grapple, the cheering crowd clustered around them, the whole scene sweeping through frame left to right as the car moves on.
Performance: The two miners each want to not back down first — play it as real escalation, the left one's temper snapping into the first shove, both grabbing on and wrestling clumsily with no technique, all anger and no skill; the surrounding crowd is loose and drunk and delighted, egging them on, hungry for the spectacle.
Camera: close-up on a long lens, dollying fast frame-left to frame-right at the car's passing speed while panning left to hold the fighting miners, the scene sweeping through frame. No cuts.
Style: smooth and methodical camera movement. 35mm film emulation, halation, bloom, shallow depth of field with hexagonal bokeh.
The amber-lit buildings behind the miners, the teal-green street haze, the warm pools of streetlight the fight tumbles through, the motion blur of the foreground as the car sweeps past.
Audio: no music, no score. The two miners' escalating shouting and cursing rising close then doppler-shifting past as the car moves, the grunts and scuffle of the grapple, the cheering and jeering of the surrounding crowd swelling and then falling away behind, the hum of the car, the pit-city street ambience, all sliding past with the car's lateral motion.
Forbidden: no white blobs on the character's faces.

How do you turn an image into a fight scene with AI?

In four steps: start from one strong image, lock how each fighter looks, frame every beat of the fight as its own image, then animate each image with one action.
  1. Step 1

    Start from one strong image.

    Upload a photo or generate a portrait in Flick with the fighter, the place and the light already right, because the video model keeps whatever the first frame holds.

    • Hannah generated both boxers this way, as portraits in the same opera hall, before any fight existed.
    Portrait of a lean 1940s boxer in white trunks and dark gloves under a spotlight beside a red velvet curtain
    Hannah L., director of Boxing at the Opera · GPT Image 2
  2. Step 2

    Lock each fighter's look.

    Turn each fighter into a character sheet with front, side and back views, so every later image of the fight keeps the same face, build and trunks.

    • Hannah's request was eight words: "create a character sheet for him, 4 angles full body".
    Character sheet of a boxer in white trunks and brown gloves, seen from four angles in front of stage lights
    Hannah L., director of Boxing at the Opera · Nano Banana Pro
  3. Step 3

    Frame each beat as an image.

    Make the wind-up, the hit and the fall as separate stills before animating anything, because a model animates one clear moment far better than it invents a whole exchange.

    Black and white close-up of a boxer with his eyes shut as a glove smashes into his face
    Hannah L., director of Boxing at the Opera · GPT Image 2
  4. Step 4

    Animate one action per image.

    Write the camera, the one action and the look, then generate; Seedance 2.5 runs on the Standard plan, and MiniMax H3 Max is free.

    • Hannah's whole prompt for this shot asked for slow motion, a locked camera, film grain and "serious fight scene".
    Two shirtless boxers in an opera house, one landing a right hand on the other's face in slow motion
    Start frame: Two shirtless boxers in an opera house, one landing a right hand on the other's face in slow motion
    Start frame
    00:05
    Hannah L., director of Boxing at the Opera · Seedance 2.5 · Image to video

What should a fight prompt say?

Name the camera first, then one action and its result, then the look and the sound, because the image already shows who is fighting and where.
  • A short prompt works when the image does the staging: Boxing at the Opera's victory shot is two sentences of action and one of film grain.
  • A long prompt earns its length when the image is a reference rather than the first frame: the Xenogenesis brawl spells out the shove, the grapple, the crowd and the camera's path.
  • The Miracle shows both ends: a stunt prompt that names every weight shift, and a prompt that hands the choreography to a depth-map video of the moves.
A boxer raises his fist in a spotlight on an opera stage while his opponent lies flat beside him
Start frame: A boxer raises his fist in a spotlight on an opera stage while his opponent lies flat beside him
Start frame
00:05

The knockout: victory, applause, a locked camera

roaring applause from the audience and stand ovation. he stands there in victory, tired and heavy breathing, the bright light shining on him. camera locked. analog film grain texture.

A static fixed film grain overlay sits on top of the entire frame, moving like a natural vintage static. The grain is staticy the entire clip. the entire scene should be a vintage analog look, an old archived film
Hannah L., director of Boxing at the Opera
A street at night with a burning car, blurred by the camera sweeping past a fight in front of lit buildings00:05

The brawl: a fight seen from a passing car

Use Miners Who Fight as the visual reference for the two arguing characters.
use Miners Womenand Miners Menas the crowd.
Use Drainage Street Angle as the visual reference for the ghetto pit-city street and the amber-lit buildings.
Single continuous shot.
Action: close-up on a long lens from the moving car's POV, the camera dollying quickly frame-left to frame-right catching a street scene in a fleeting passing glance. Two arguing characters stand directly in the front of the buildings, out of the streets, next to the burning car, face each other in profile, one getting up in the other's face, arguing heatedly, chests nearly touching, on the verge of throwing punches. Then the left character shoves and swings hits the other character in the eye. Then they grab each other, grappling in a messy, real, unchoreographed tussle, staggering and wrenching at each other's clothes, no clean martial choreography, just two furious men wrestling on the street. Around them, the crowd gathers and cheer them on, throwing up their arms and shouting. 
Blocking: The two characters are standing on the sidewalk of . They face each other in profile as the camera dollies right at car-speed to hold them in frame just long enough to catch the shove and grapple, the cheering crowd clustered around them, the whole scene sweeping through frame left to right as the car moves on.
Performance: The two miners each want to not back down first — play it as real escalation, the left one's temper snapping into the first shove, both grabbing on and wrestling clumsily with no technique, all anger and no skill; the surrounding crowd is loose and drunk and delighted, egging them on, hungry for the spectacle.
Camera: close-up on a long lens, dollying fast frame-left to frame-right at the car's passing speed, the scene sweeping through frame. No cuts.
Style: smooth and methodical camera movement. 35mm film emulation, halation, bloom, shallow depth of field with hexagonal bokeh.
The amber-lit buildings behind the miners, the teal-green street haze, the warm pools of streetlight the fight tumbles through, the motion blur of the foreground as the car sweeps past.
Audio: no music, no score. The two miners' escalating shouting and cursing rising close then doppler-shifting past as the car moves, the grunts and scuffle of the grapple, the cheering and jeering of the surrounding crowd swelling and then falling away behind, the hum of the car, the pit-city street ambience, all sliding past with the car's lateral motion.
Forbidden: no white blobs on the character's faces.
Gertywood, director of Xenogenesis
A man in a torn ivory suit rolls across a wet tower top at night, about to tackle an armored knight
Start frame: A man in a torn ivory suit rolls across a wet tower top at night, about to tackle an armored knight
Start frame
00:05

The tackle: one stunt, every weight shift spelled out

Non-graphic controlled fantasy stunt for a 1980s practical-action film. From the start frame, the Runner finishes the same roll after the blade misses, plants one hand and one foot on the wet circular tower, drives his body weight back toward the black-armored Sentinel, tackles him at the torso, and lands on top wrestling for the same double-ended polearm, ending exactly as the end frame. Move through the intermediate blocking of the Runner reengaging the Sentinel, then tackling him, then both pairs of hands wrestling for the polearm. Make every body-weight transfer readable and grounded; no teleporting, no sliding, no standing jump, no extra limbs, no extra weapons. Preserve exact Runner face, wet dark hair, glossy ivory tailored suit, Sentinel black armor and orange visor, the polearm, tower geography, camera axis. Restrained 2.39:1 framing, rain, storm sky, perimeter lights, amber doorway glow.
Sam Wasserman, director of The Miracle
Low view across a rain-soaked circular tower top under a storm sky, an armored boot at the edge of frame00:15

The duel: a depth-map video drives the choreography

[Intent & Subject] Create exactly 15 seconds, 21:9, native 2K. One Runner and one Sentinel only. Runner: single dark-haired man in soaked torn pearl-white 1980s suit with medallion, unarmed. Sentinel: single towering black-armored knight with narrow red-orange visor and exactly one double-ended poleaxe. [Motion Source] @Video1 is a depth-map video. Use its moving depth silhouettes as the exclusive choreography, blocking, attack timing and camera motion; it is not the visual look. Retarget the principal unarmed depth figure to Runner and principal armed depth figure to Sentinel, preserving rolls, evasions, throws, strikes, weapon paths, weight shifts and contact beats. Omit incidental figures. [Environment & Style] Rain-soaked black circular tower summit, wet concentric rings, intact sheer rim over storm city, central dark tower and broad warm golden doorway. Photoreal 1980s 35mm anamorphic action-fantasy, practical rain, steel-blue lightning, tungsten gold, grain, halation, slight gate weave. [Cinematography] Dynamic tight coverage: boots, hands on poleaxe shaft, gauntlet grip, blade near face, visor glow, wet eye, torn sleeve and impacts; wides only for readable rolls, throws and edge danger. Real mass, no floaty physics. [Audio] No music or dialogue; rain, wind, armor, boots, poleaxe, impacts and breath. [Continuity & Exclusions] Exactly two living figures total in every frame, exactly one Runner plus exactly one Sentinel. No clones, duplicates, twins, crowd, background fighters, reflections, silhouettes, extra limbs, extra weapons or identity swaps. Runner never gains a weapon; Sentinel alone owns the poleaxe.
Sam Wasserman, director of The Miracle
What each part of a fight prompt does, with words copied from the Boxing at the Opera and Xenogenesis prompts
Part of the promptWhat it fixesWritten this way on the canvas
CameraHow the shot moves, so the eye stays on the fighterscamera locked.
SpeedHow long a hit lasts on screenslow motion.
ActionThe one thing that happens in the clipmid action shot of him getting punched.
LookThe texture that ties every shot togetheranalog film grain texture.
Fighting styleHow the fight looks, clean or messyno clean martial choreography, just two furious men wrestling on the street
SoundWhat you hear, and what you do notAudio: no music, no score.

Why make a fight scene in Flick instead of a one-click fight app?

A one-click app turns one photo into one clip, while Flick lets you build a fight scene from many images and clips on one canvas, with the same fighters in every shot.
  • Image first

    Every clip starts from a still you upload or make, so the fighters, the place and the light are set before anything moves.

  • Same fighters, every shot

    Character sheets and reference images keep each fighter's face and costume from the stare-down to the knockout.

  • Video models side by side

    Try Seedance 2.5, Kling 3.0 Pro or MiniMax H3 Max on the same image and keep the take that hits hardest.

  • A scene, not a clip

    Images, clips and cuts sit on one canvas, so a fight grows beat by beat into a scene with a winner and a crowd.

What do people ask about AI fight generators?

What is an AI fight generator?

An AI fight generator is a video tool that turns an image or a text description into a fight clip, such as a boxing match, a brawl or a sword fight. In Flick, the image becomes the first frame and the prompt describes the action.

Can I turn a photo into a fight video?

Yes. Upload the photo to a Flick canvas, connect it to an image to video model as the first frame, and describe the fight in the prompt. Use photos of people who have agreed to it.

Is Flick's AI fight generator free?

Yes, you can start free: you get 100 credits when you sign up, 200 more for joining the Flick Discord, and 100 credits every week. MiniMax H3 Max video is on the free plan; Seedance 2.5 needs the Standard plan ($20 a month).

How long can an AI fight clip be?

Seedance 2.5 makes clips of 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 or 30 seconds in Flick. Short clips of one action each, cut together, usually read better than one long clip, because each hit stays clear.

How do I keep the fighters looking the same in every shot?

Make a character sheet of each fighter, with front, side and back views, and use it in every image and clip of the fight. Starting each clip from an image of the same fighters keeps faces and costumes steady too.

Can AI make a sword fight or a martial arts fight?

Yes. Describe each fighter's weapon and how they move, then break the fight into timed beats. Flick's guide on how to choreograph a fight scene with AI follows a 30 second spear and sword duel step by step.

Who owns the fight videos I make?

You do. Content you generate on Flick is yours for personal and commercial projects. See the terms of use for detail.

More on fight scenes and image to video in Flick

Films on this page

Turn your image into a fight scene

Start from a photo, a portrait or a character sheet, write the action, and let Flick animate it.

Make a fight scene in Flick

Try for free, no credit card required.

Ask AI about this page:ChatGPTPerplexityClaude