Water Sorceress AI Video Prompt: Fight in the Rain
A water sorceress AI video prompt for Seedance 2.5. You rise from a fountain, beat two swordsmen and call a giant dog made of water. 25 seconds, 9:16.
Have it running in three steps
- 1
- 2Open OpenArt
- 3Paste it and generate
What you'll need
30 min- OpenArt
This water sorceress AI video prompt makes you the hero of a night fight in a castle courtyard. Heavy rain falls on the old stones. You burst up out of a fountain in a long red leather coat, and two men with swords run at you.
You stop them with water. A wall of water knocks one sword away, and a ring of water throws both men back. Then a giant dog made of clear water rises behind you, jumps over your head and smashes them into the ground. At the end you raise one hand, and the dog falls apart into rain.
It is 25 seconds long, with fast cuts like a film fight scene. You need one clear photo of yourself. This page gives you two prompts. The first makes a character sheet of you in the outfit. The second makes the video.
The whole process
- Pick one clear photo of your face and body, in good light.
- Open OpenArt and make a character sheet with GPT Image 2.5 and the first prompt below.
- Go to video and pick Seedance 2.5.
- Add the character sheet as the only reference image.
- Set 25 seconds and 9:16. Paste the video prompt and generate.
What happens in 25 seconds
| Time | What happens |
|---|---|
| 0–3s | The fountain explodes. She stands up in the water, soaked, and looks at the camera |
| 3–5s | The first man runs in from the left and swings his sword. She ducks under it |
| 5–9s | Her hair spins, and the second man jumps onto the fountain edge. A wall of water hits his sabre and he falls |
| 9–11s | Both men run at her. She opens her arms, and a ring of water throws them across the stones |
| 11–13s | Close-up: her hands go into the water, and it starts to move |
| 13–16s | A giant clear water dog rises behind her, shows its teeth and jumps |
| 16–19s | It lands on both men. They fly through the air and lie still on the ground |
| 19–23s | She steps out of the fountain and walks to the camera. The dog walks behind her |
| 23–25s | She raises one hand. The dog falls apart into rain, and she stands alone |
Make the character sheet
The coat must look the same in every cut: when she ducks, spins and walks. One photo does not show the coat from the back or the side. So first make a character sheet: a close-up of the face and four full-body views.
Open OpenArt, pick GPT Image 2.5 and add your photo as the reference. Set the shape to 16:9 and paste this prompt. Your face, hair and body come from the photo. The outfit is written in the prompt, so your own clothes do not matter.
Character reference sheet, 16:9 landscape, plain mid-grey studio background, soft even front light with gentle rim light, natural skin texture, sharp fabric detail. One woman only — face, hair and body the same as in the reference image; ignore the reference image’s clothing, background and lighting. Layout: left third — a close-up of her head and shoulders, front-facing, calm intense gaze, collar and top of the bodice visible. Right two thirds — four full-body views in one row, same height and scale, standing in a neutral pose, arms slightly away from the body: front view, left side view, back view, right side view. Outfit: a long deep crimson leather coat with a soft matte, slightly worn finish, fitted through the shoulders and waist, long fitted sleeves, a wide seamed waist panel, open at the front from the waist down, high-low hem — mid-thigh at the front edges, mid-calf at the back, flaring as it falls. Under it a dark burgundy leather corset bodice with black front lacing and a V neckline. Black fitted trousers, knee-high matte black leather boots with a buckle strap below the knee. Short fingerless crimson leather gloves with a wrist strap. The outfit is identical in all five views. No props, no glow, no jewelry, no text, no labels.
This is the sheet we used. A close-up of the face on the left, then the front, left side, back and right side.
Look at the back view before you go on. The coat should be long at the back and short at the front. If the hem is wrong on the sheet, it will be wrong in the video too.
Set it up and generate
In OpenArt, go to video and pick Seedance 2.5.
Add the character sheet as the only reference. It becomes @Image1 in the
prompt. Set the length to 25 seconds and the shape to 9:16. Paste the
prompt below as it is.
The prompt only says her look comes from @Image1. It does not describe her
face in words. The two men and the water dog are written out in full, so the
model does not need any other picture.
NO music. Diegetic sound only.
[LOOK — LOCKED, APPLIES TO EVERY SHOT] ARRI Alexa 35, Cooke S4/i primes: natural skin, soft highlight roll-off, creamy bokeh on lit windows. Fine film grain, deep blacks, moderate contrast. Storm Night grade: cold slate-blue stone and rain, warm amber windows as the only warm light, the crimson coat as the only saturated colour. Torrential rain in every shot — visible streaks, splashes on stone, everyone soaked. Fight shots: handheld with energetic shake and whip pans. Opening and closing shots: steady. Edited as an action scene with hard cuts, 1–3 seconds per shot during the fight.
[References] The woman corresponds to @Image1. Use only her face, hair, body, outfit and crimson coat, the same as in the reference image. Do not use the background, lighting or pose of @Image1.
[Cast — exactly three people] The woman. Attacker A: short dark hair, black leather jacket, longsword, always from frame left. Attacker B: shaved head, dark grey hooded coat, curved sabre, always from frame right. No other people at any time.
[Generation Goal] A 25-second dark fantasy action scene at night in a gothic castle courtyard in torrential rain. A sorceress bursts out of a stone fountain, fights off two swordsmen with water magic, summons a giant dog made of water that smashes them down, then dissolves it into rain.
[Stage 1, 0–3 s] Shot 1: low angle wide. The fountain erupts and she bursts up out of the water to standing in one sharp motion, water pouring off her coat. Camera holds, then a short push-in as she lifts her head. End state: she stands knee-deep in the basin, eyes locked on the attackers.
[Stage 2, 3–11 s] Continue from the previous stage: same woman, same coat, same fountain. Shot 2 (3–5 s): handheld over Attacker A’s shoulder. He sprints in from the left and swings at her head; she ducks under the blade, spray flying off her hair. Shot 3 (5–7 s): close-up on her face as she spins toward the right; whip pan to Attacker B leaping onto the fountain rim, sabre raised overhead. Shot 4 (7–9 s): medium low angle. She thrusts her palm up; a wall of water erupts from the basin, his sabre crashes into it and is knocked aside, he stumbles backward off the rim. Speed ramp: slow motion on the impact, back to real time. Shot 5 (9–11 s): wide. Both men charge together from both sides; she sweeps both arms outward and a ring of water blasts them back across the stones. End state: both attackers on the cobblestones, crouched, swords ready, breathing hard; she stands in the basin.
[Stage 3, 11–15 s] Continue from the previous stage: positions unchanged. Shot 6 (11–13 s): extreme close-up. Her hands plunge into the water; it starts to churn and climb. Shot 7 (13–15 s): low angle from behind the attackers. A giant dog made entirely of clear water rises out of the fountain behind her — German Shepherd shape, upright ears, long muzzle, four legs, one tail, pale glowing eyes, twice the height of a man. Its body is transparent rippling water with no fur; the courtyard and rain are visible through it. It bares its water fangs. Camera tilts up with it. End state: the water dog towers behind her, crouched to spring.
[Stage 4, 15–19 s] Shot 8 (15–17 s): the dog leaps over her head toward camera; both attackers raise their swords. Camera tracks backward. Shot 9 (17–19 s): its front paws crash down on both men and burst into a wave of spray; both are thrown through the air and skid across the wet stones, swords spinning away. Speed ramp: slow motion as they fly, real time as they land. End state: both men lie still on the ground; the water dog re-forms between them.
[Stage 5, 19–25 s] Continue from the previous stage: attackers stay on the ground until the end. Shot 10 (19–23 s): steady camera. She steps over the fountain rim and walks toward camera; the water dog pads behind her. Camera pulls back. Shot 11 (23–25 s): wide. She raises one open hand; the dog collapses into fine mist that falls as rain. End state: she stands alone in the rain, both attackers lying on the cobblestones.
[Maintain Consistency] Exactly three people. Her face, hair, outfit and crimson coat never change. The dog is always transparent water with dog anatomy — no fur, never a bear or serpent, one head. Magic is water only: no light beams, no energy bolts, no fog. Swords never pass through bodies. The fountain shape never changes.
Audio: NO music. <heavy rain on stone>, <water erupting>, <blades slicing air>, <sabre slamming into water>, <boots splashing>, <deep watery growl>, <crashing wave and bodies hitting wet stone>, <soft hiss of mist>. Final seconds: rain alone.
Why this prompt works
The look is set once, at the top, for every shot. Cold blue stone, warm yellow windows and one bright colour: the red coat. So your eye always finds her, even in the fast fight cuts.
Each man has his own side of the frame. The first one always comes from the left, the second from the right. That keeps the fight easy to follow, and the model does not mix up the two men.
Every stage ends with an “End state”. It says where everyone is when the stage is over. The next stage starts from that point, so the men do not jump back to their feet or swap places between cuts.
The magic is water only. The prompt says no light beams, no energy bolts and no fog. Fight scenes pull models toward glowing blasts, and this line keeps every attack wet and real.
The music is turned off. You hear only rain, swords, splashes and a deep growl, and the last seconds are just rain. That makes the quiet ending feel heavy.
Where it goes wrong
| What you got | The fix |
|---|---|
| The dog looks like a bear or a lion | Keep “German Shepherd shape, upright ears, long muzzle” and “never a bear or serpent” |
| The dog is grey with fur, not clear water | Keep “transparent rippling water with no fur; the courtyard and rain are visible through it” |
| A third man or a crowd shows up | Keep “Exactly three people” and “No other people at any time” |
| Blue light beams come out of her hands | Keep “Magic is water only: no light beams, no energy bolts, no fog” |
| The men stand up again at the end | Keep “attackers stay on the ground until the end” in Stage 5 |
| Her coat changes shape or colour | Make the sheet again with a clear back view. Keep “Her face, hair, outfit and crimson coat never change” |
| Background music plays | Keep “NO music. Diegetic sound only.” as the first line |
| A grey studio wall shows up | Keep “Do not use the background, lighting or pose of @Image1” |
| It is one long shot with no cuts | Keep “Edited as an action scene with hard cuts, 1–3 seconds per shot during the fight” |
Questions people ask
Which AI model makes this video?
We made it with Seedance 2.5 on OpenArt, 25 seconds, with one reference image. The character sheet was made with GPT Image 2.5 on the same platform.
Can I put myself in it?
Yes. Make the character sheet from your own photo. The video copies your face and hair from the sheet.
Does it work for a man?
Yes, with small changes. In both prompts change “woman” to “man” and “she” and “her” to “he” and “his”. In the sheet prompt, swap the corset for a fitted leather vest.
Is it vertical?
Yes, 9:16, ready for Reels, TikTok and Shorts.
Is it free?
The prompts are free. OpenArt runs on credits, so the sheet and the video use credits from your account.
Can I change the water dog to another animal?
Yes. Change the dog lines in Stage 3 and in “Maintain Consistency”. Describe the new animal just as clearly: its shape, legs, tail and what it must never turn into.
Ready? Open OpenArt, make your sheet and paste the video prompt. For more scenes like this, see all our video prompts.
Why this workflow runs here
Many models, one account
Free credits every day
Compare before you commit
Make yours in the next 30 min
You have the whole prompt and every setting. The only thing left is an account, and that takes a minute.
- Free to start
- Prompts ready to paste
- Upgrade only if you want more
Referral link — if you subscribe I earn a commission, at no extra cost to you. More on that.
Related prompts
Cinematic Free Action Scene AI Video Prompt: Bike Jump and Fight
Action scene AI video prompt for Seedance 2.5. You jump a futuristic bike off a ramp, land in fire and fight on a burning highway. 15 seconds, one photo.
Cinematic Free Anime Borscht — The Whole Recipe in One 25-Second Video
One prompt cooks Ukrainian red borscht from raw meat to a served bowl, in glossy anime style. Six stages, hands and food only, vertical for Reels.
Cinematic Free APT AI Prompt — A Dog and a Cat Do the APT. Dance
Free APT AI prompt for Seedance 2.5. A dog sings and a cat plays drums in the APT. dance. Download the clip, make two sheets, swap in your own pets.