How to Make Consistent AI Characters for Free (Same Face in Every Shot)

Making a consistent AI character is the hardest part of AI filmmaking. Every AI creator hits the same wall: you make a character, and in the next shot the face changes. So does the outfit, the gun and even the helicopter. I wanted to see if free tools could fix that, so I picked a hard test: two US Army Special Forces soldiers and a Black Hawk helicopter, landing, jumping out and in a firefight, with the same faces in every shot.

The 7 steps at a glance:

1. Get a character sheet template from Claude.
2. Create a character sheet for each person in ChatGPT.
3. Make a separate sheet for every vehicle or prop.
4. Combine the sheets into scene images.
5. Write a fixed character lock for everyone.
6. Add the sheets and locks as references in Google Flow and generate.
7. Test Dreamina (Seedance) for action shots, and trim small mistakes instead of regenerating.

Last updated: November 9, 2026. Tested on my own free accounts.

Watch the Video Tutorial

The whole workflow, including my mistakes and failed generations, is in this video:

https://youtu.be/WBKUfAKEyVU

New to AI characters? Start with my first guide: How I made a realistic AI character with 6 free AI tools.

The Free Tools I Used

Tool	What it did	Free limit (Nov 2026, my account)
Claude	Character template, character locks, video prompts	Free plan
ChatGPT	Character sheets and scene images	A few images a day (limit resets)
Google Flow	Talking video clips with the same faces	50 credits a day (one clip = 24 credits)
Dreamina (Seedance 2.0 Fast)	Action clip (firefight)	120 credits a day (5 s at 1080p = 70 credits)

Free limits change often and differ by account and country, so check them before you start.

Why Your AI Character Is Not Consistent

AI video tools don’t remember your character from one clip to the next. Every generation starts fresh, so if your prompt only says “a soldier with a beard”, the model invents a new soldier each time. The fix is to give it the same picture and the same words every single time: a character sheet as the image reference, and a character lock as the text reference.

Step 1: Get a Character Sheet Template from Claude

I created a Claude project called AI Film and told it to act as my film expert. Then I asked it to turn my rough character sheet idea into a clean, structured template. You can copy the template below and fill in the brackets for any character.

Create a single landscape image (16:9) of a professional character concept art sheet on a pure white background.

CHARACTER: [age, gender, ethnicity, height/build, face shape, hair style and color, eye color, skin details, facial hair, expression]
OUTFIT: [full outfit/armor description, materials, colors, wear and tear]
EQUIPMENT: [weapons/gear and where they are carried — holster, back, belt]
UNIQUE DETAILS: [patches, tattoos, scars, insignia, jewelry, dog tags]

LAYOUT (follow exactly):
- LEFT TWO-THIRDS: full-body turnaround of the same character standing in a neutral A-pose, shown three times side by side at the same scale and height — front view (left), side profile view (middle), back view (right). Head to toe fully visible, feet on the same ground line.
- TOP RIGHT: a 3x2 grid of 6 head-and-shoulders portraits of the same face — front, 3/4 left, 3/4 right, left profile, right profile, slight upward angle. Neutral expression, consistent lighting.
- BOTTOM RIGHT: a 2x2 grid of 4 close-up detail shots: [DETAIL 1], [DETAIL 2], [DETAIL 3], [DETAIL 4].
- Thin even white gaps between all panels, clean and organized.

STYLE: hyper-realistic, AAA game character design, photoreal skin and fabric textures, soft studio lighting with gentle shadows, sharp focus, ultra-detailed, same identity and outfit in every view.

STRICT: absolutely no text, no letters, no labels, no numbers, no arrows, no captions, no watermarks, no logos anywhere in the image.

Step 2: Create the Character Sheets in ChatGPT

A character sheet shows the character from the front, side and back, plus six face angles and four detail close-ups. This is the image every video tool will copy, so take your time here.

Male operator: Sgt. Ryan Cole

Create a single landscape image (16:9) of a professional character concept art sheet on a pure white background.

CHARACTER: 34-year-old American man, 6'1", athletic muscular build, square jaw, short dark brown hair faded on the sides, short trimmed full beard, hazel eyes, light tan skin with small scar on left eyebrow, calm focused expression.
OUTFIT: Multicam combat shirt with sleeves rolled to forearm, multicam combat pants with knee pads, coyote-brown plate carrier with magazine pouches, tan tactical gloves, coyote-brown combat boots, olive baseball cap.
EQUIPMENT: M4 carbine on a two-point sling across the chest, Glock pistol in a thigh holster, radio on the left shoulder strap, small assault backpack, fixed-blade knife on the belt.
UNIQUE DETAILS: American flag patch in subdued colors on the right shoulder, dog tags on a chain, worn leather watch, light dust and wear on the gear.

LAYOUT (follow exactly):
- LEFT TWO-THIRDS: full-body turnaround of the same character standing in a neutral A-pose, shown three times side by side at the same scale and height — front view (left), side profile view (middle), back view (right). Head to toe fully visible, feet on the same ground line.
- TOP RIGHT: a 3x2 grid of 6 head-and-shoulders portraits of the same face — front, 3/4 left, 3/4 right, left profile, right profile, slight upward angle. Neutral expression, consistent lighting.
- BOTTOM RIGHT: a 2x2 grid of 4 close-up detail shots: the flag shoulder patch, the dog tags and chain, the M4 carbine receiver, the worn combat boots.
- Thin even white gaps between all panels, clean and organized.

STYLE: hyper-realistic, AAA game character design, photoreal skin and fabric textures, soft studio lighting with gentle shadows, sharp focus, ultra-detailed, same identity and outfit in every view.

STRICT: absolutely no text, no letters, no labels, no numbers, no arrows, no captions, no words on the patch, no watermarks, no logos anywhere in the image.

Tip: after the sheet, ask ChatGPT for two more images: a clean full-body turnaround and a face-only portrait. The face photo becomes the character’s “ID card” in Flow and Dreamina.

Female operator: Sgt. Ava Reed

The first face ChatGPT gave me didn’t feel right, so I simply asked it to change the face. Fix the face now, because after this step it’s locked.

Create a single landscape image (16:9) of a professional character concept art sheet on a pure white background.

CHARACTER: 26-year-old American woman, 5'8", fit athletic build with strong shoulders, oval face with defined cheekbones, dark blonde hair tied in a tight low bun with a few loose strands, green-hazel eyes, fair skin with light freckles across the nose, small healed scar on the chin, confident calm and focused expression, no makeup.
OUTFIT: Multicam combat shirt with sleeves rolled to the forearm, multicam combat pants with knee pads, coyote-brown plate carrier with magazine pouches, tan tactical gloves, coyote-brown combat boots, olive baseball cap, practical realistic military fit (not stylized, not revealing).
EQUIPMENT: M4 carbine on a two-point sling across the chest, Glock pistol in a thigh holster, radio on the left shoulder strap, small assault backpack, fixed-blade knife on the belt, tactical headset around the neck.
UNIQUE DETAILS: subdued American flag patch on the right shoulder, dog tags on a chain, black digital watch, green Special Forces style beret clipped to the backpack, light dust and wear on the gear.

LAYOUT (follow exactly):
- LEFT TWO-THIRDS: full-body turnaround of the same character standing in a neutral A-pose, shown three times side by side at the same scale and height — front view (left), side profile view (middle), back view (right). Head to toe fully visible, feet on the same ground line.
- TOP RIGHT: a 3x2 grid of 6 head-and-shoulders portraits of the same face (without cap) — front, 3/4 left, 3/4 right, left profile, right profile, slight upward angle. Neutral expression, consistent lighting.
- BOTTOM RIGHT: a 2x2 grid of 4 close-up detail shots: the flag shoulder patch, the dog tags and chain, the M4 carbine receiver, the green beret clipped to the backpack.
- Thin even white gaps between all panels, clean and organized.

STYLE: hyper-realistic, AAA game character design, photoreal skin and fabric textures, soft studio lighting with gentle shadows, sharp focus, ultra-detailed, same identity and outfit in every view.

STRICT: absolutely no text, no letters, no labels, no numbers, no arrows, no captions, no words on the patch, no watermarks, no logos anywhere in the image.

Step 3: Make a Sheet for the Vehicle Too

Vehicles change even more than faces. So the Black Hawk gets its own sheet: four views, cockpit, cabin, minigun, refueling probe and close-ups of the sensor, tail rotor, exhaust and wheels.

Create a single landscape image (16:9) of a professional vehicle concept art sheet on a pure white background.

VEHICLE: US Army MH-60M Black Hawk special operations helicopter. Four-blade main rotor, four-blade tail rotor on the left side of the tail, long refueling probe extending forward from the right side of the nose, twin turboshaft engines on top, forward-looking infrared (FLIR) sensor ball under the nose, external fuel tanks on stub wings, sliding cabin doors open on both sides with M134 minigun door mounts, fast-rope bars above the doors.
PAINT & SURFACE: matte dark grey-black special operations paint, slightly faded and weathered, light oil streaks behind the engine exhausts, dust on the landing gear and lower fuselage, scuffed edges on the doors, realistic rivets and panel lines.
UNIQUE DETAILS: small subdued low-visibility American flag on the tail, dark-tinted cockpit windows, infrared exhaust suppressors, wire-strike cutters above and below the cockpit, two small antennas on the tail boom.

LAYOUT (follow exactly):
- LEFT TWO-THIRDS: the same helicopter shown four times at the same scale — left side view (top left), front view (top right of this area), rear view (bottom left), top-down view with rotors spread (bottom right of this area). Rotors still, full aircraft visible, all on clean white.
- TOP RIGHT: a 2x2 grid showing the cockpit interior with glass displays and two pilot seats, the open cabin interior with troop seats and fast-rope gear, the door-mounted minigun, and the refueling probe on the nose.
- BOTTOM RIGHT: a 2x2 grid of close-ups showing the FLIR sensor ball, the tail rotor, the engine exhaust area, and the main landing wheel.
- Thin even white gaps between all panels, clean and organized.

STYLE: hyper-realistic, AAA game vehicle asset, photoreal metal, paint and glass materials, soft studio lighting with gentle shadows, sharp focus, ultra-detailed, identical aircraft design and paint in every view.

STRICT: absolutely no text, no letters, no tail numbers, no serial numbers, no "US ARMY" lettering, no labels, no arrows, no captions, no watermarks, no logos anywhere in the image.

Right here I hit ChatGPT’s free image limit. It’s normal. The limit resets, so come back later and continue in the same chat (your reference images stay there).

Step 4: Build the World with Scene Images

Now attach all three sheets and ask for full scenes. These images show the video tool what the world looks like: the people, the aircraft, the light and the location.

Scene A: walking away from the helicopter

Attach: male sheet, female sheet, helicopter sheet.

Using the three attached reference sheets, create a single hyper-realistic cinematic photo (16:9). Keep both characters' faces, outfits and gear EXACTLY as in their reference sheets, and keep the helicopter EXACTLY as in its reference sheet.

SCENE: a remote desert forward operating base at golden hour, sun low on the horizon behind the mountains, warm orange light with long shadows, light dust kicked up by the rotor wash. The MH-60M Black Hawk sits on a gravel landing zone in the mid-ground, rotors slowly spinning with slight motion blur, cabin door open, a pilot faintly visible in the cockpit.

CHARACTERS (foreground, walking toward camera, side by side):
- LEFT: the 34-year-old male operator from reference sheet 1, M4 carbine held low-ready, looking slightly to his right, serious focused expression.
- RIGHT: the 26-year-old female operator from reference sheet 2, M4 carbine held low-ready, tactical headset on, a few strands of hair moving in the rotor wind, calm confident expression looking straight ahead.
Both are full body, walking in step, with dust swirling around their boots.

CAMERA: low angle at waist height, 35mm lens, shallow depth of field with the characters sharp and the helicopter slightly soft, realistic film grain, natural color grading like a modern war film.

STYLE: photorealistic, cinematic, ultra-detailed skin, fabric and metal textures, realistic military proportions and gear.

STRICT: no text, no letters, no tail numbers, no lettering on patches or the helicopter, no captions, no watermarks, no logos.

Scene B: the moment of landing

Attach: scene image A (for the characters) and the helicopter image.

Using the two attached reference images, create a single hyper-realistic cinematic photo (16:9).
Keep the MALE and FEMALE operators EXACTLY as in reference image 1 (same faces, hair, beard, outfits, gear), and keep the helicopter EXACTLY as in reference image 2 (same matte dark grey-black weathered paint, refueling probe on the nose, FLIR sensor ball under the nose, external fuel tank on the stub wing, door-mounted minigun, open sliding cabin door).

MOMENT: the exact moment of landing. The MH-60M Black Hawk is in a landing flare about 1 meter above the gravel landing zone, nose slightly raised, main wheels just about to touch the ground. Rotor wash blasts a huge cloud of desert dust and small stones outward in a circle. Rotors spinning fast with strong motion blur.

CHARACTERS (seated in the open left cabin doorway, clearly visible, facing the camera side):
- FRONT (closer to cockpit): the male operator from reference 1, bearded face, olive cap, multicam uniform, coyote plate carrier, dog tags, seated on the cabin floor edge with boots resting on the step, M4 carbine held vertical between his knees, gripping the door frame with his left hand, eyes scanning the ground, serious focused expression.
- BEHIND HIM (toward the tail): the female operator from reference 1, dark blonde hair in a messy low bun with loose strands whipping in the rotor wind, tactical headset on her ears, light freckles, same multicam uniform and plate carrier, seated on the cabin floor edge with one boot on the step, M4 carbine across her lap, ready to jump out, calm confident expression looking forward.
Both are wearing gloves, both have a safety tether clipped to the cabin floor, and their clothing ripples in the wind. A pilot is faintly visible in the cockpit.

ENVIRONMENT: same remote desert forward operating base as the references, wooden guard tower and stacked HESCO barriers in the background, green ammo crates on the left, mountains on the horizon, golden-hour sun low behind the mountains, warm orange backlight, sun rays cutting through the dust.

CAMERA: ground-level shot from the left side at a slight 3/4 front angle, 35mm lens, the whole helicopter in frame, both characters sharp and clearly recognizable, dust in the foreground slightly out of focus, realistic film grain, modern war-film color grading.

STYLE: photorealistic, cinematic, ultra-detailed skin, fabric, metal and dust textures, realistic military proportions and gear.

STRICT: same two people as the reference, no extra soldiers in the doorway, no text, no letters, no tail numbers, no lettering on patches or the helicopter, no captions, no watermarks, no logos.

Step 5: Write a Character Lock for Everyone

This is the pro trick. A character lock is a fixed block of text that describes age, face, hair, body, outfit and even the voice. The helicopter gets one too, down to the sound of its rotors. Never change a single word. Paste the exact same lock into every clip.

Lock: Ryan Cole

CHARACTER LOCK — SGT. RYAN COLE (male operator)
AGE: 34 years old, mature rugged face, never younger or older.
FACE: square jaw, strong straight nose, deep-set hazel eyes, thick dark eyebrows, small scar through the left eyebrow, short dense dark brown full beard neatly trimmed, light tan weathered skin with fine sun lines at eyes.
HAIR: short dark brown hair, faded on the sides, mostly hidden under an olive-tan baseball cap.
BODY: 6'1" (185 cm), broad shoulders, muscular athletic build, thick forearms, upright confident posture, heavy steady walk.
OUTFIT (never changes): multicam combat shirt with sleeves rolled to the forearm, multicam pants with coyote knee pads, coyote-brown plate carrier with magazine pouches and radio on the left shoulder strap, subdued American flag patches on both shoulders, silver dog tags hanging outside the shirt, tan tactical gloves, black digital watch, coyote-brown combat boots, Glock pistol in a right thigh holster, black M4 carbine on a sling.
VOICE: deep low baritone, calm and controlled, slightly gravelly, General American accent (Midwest), short clipped military sentences, speaks slowly with authority, never shouts unless in combat.
PERSONALITY ON SCREEN: serious, protective, few words, scans surroundings constantly.

Lock: Ava Reed

CHARACTER LOCK — SGT. AVA REED (female operator)
AGE: 26 years old, youthful but strong face, never younger or older.
FACE: oval face with defined cheekbones, green-hazel eyes, straight medium eyebrows, light freckles across the nose and cheeks, natural full lips, small healed scar on the chin, fair skin with a light sun tan, no makeup.
HAIR: dark blonde to light brown hair tied in a messy low bun, several loose wavy strands framing the face.
BODY: 5'8" (173 cm), fit athletic build, toned arms and strong shoulders, light quick confident stride.
OUTFIT (never changes): multicam combat shirt with sleeves rolled to the forearm, multicam pants with coyote knee pads, coyote-brown plate carrier with magazine pouches and radio on the left shoulder strap, subdued American flag patches on both shoulders, dog tags, olive tactical headset over the ears, tan tactical gloves, black digital watch, coyote-brown combat boots, Glock pistol in a right thigh holster, black M4 carbine on a sling. Practical realistic military fit, never stylized or revealing.
VOICE: clear medium-pitched female voice, slightly husky, calm and confident, General American accent (West Coast), quick precise delivery, steady even under pressure, occasional dry humor.
PERSONALITY ON SCREEN: sharp, alert, focused, quick to act.

Lock: Shadow 6 (the helicopter)

VEHICLE LOCK — "SHADOW 6" (MH-60M Black Hawk)
DESIGN (never changes): US Army MH-60M Black Hawk special operations helicopter, matte dark grey-black paint heavily weathered with chipped edges and dust stains, long refueling probe extending forward from the right side of the nose, FLIR sensor ball under the nose, external fuel tank on the stub wing, four-blade main rotor, tail rotor on the left of the tail, sliding cabin doors with door-mounted M134 minigun, dark-tinted cockpit glass, two pilots in olive flight helmets.
SOUND: deep rhythmic rotor thump "whop-whop-whop", high-pitched turbine engine whine underneath, heavy wind roar and gravel spray during landing and takeoff, sound gets louder and lower as it approaches.
NO TEXT: no tail numbers, no lettering, no logos on the aircraft.

Step 6: Animate Your Consistent AI Character in Google Flow

1. Open Google Flow and create a new project. Set the ratio to 16:9.
2. Open the instructions / ingredients panel and add each reference image: Ryan’s face photo with his lock, Ava’s face photo with hers, and the helicopter image with its lock.
3. Give each reference a name you can tag: @RYAN_REF, @AVA_REF, @SHADOW6_REF.
4. Paste the scene prompt below. Tag every character with @ so Flow uses the right reference.
5. Flow asks before it spends credits. This 8-second clip cost 24 credits.

@RYAN_REF  @AVA_REF  @SHADOW6_REF
Use the three attached reference images. Keep every face, body, outfit and the helicopter EXACTLY as in the references. Same people, same aircraft, no changes.

[PASTE RYAN LOCK] [PASTE AVA LOCK] [PASTE SHADOW 6 LOCK]

BACKGROUND / LOCATION
Remote desert forward operating base at golden hour. Flat gravel landing zone with small stones. Background left: tall wooden guard tower and a long row of tan HESCO barrier walls with a desert camouflage net. Foreground left: stacked olive-green ammo crates. Behind: dry brown rocky mountain range on the horizon. The sun sits low on the LEFT, just above the mountains, golden-orange sky with soft scattered clouds, strong warm backlight, long shadows stretching toward the camera, sun rays cutting through floating dust.

ACTION TIMELINE (8 seconds, one continuous shot)
0–3 s — LANDING: @SHADOW6_REF descends into frame from upper right, nose slightly raised in a landing flare, rotor wash blasting a huge circular cloud of dust and small stones outward. @RYAN_REF and @AVA_REF sit in the open cabin doorway, boots on the step, rifles in hand, Ava's loose hair strands whipping in the wind. The wheels touch down hard on the gravel and the suspension compresses.
3–5 s — JUMP: Ryan jumps out first and lands in a slight crouch, then Ava jumps right behind him. Both land firmly, dust bursting around their boots, rifles held low-ready.
5–8 s — WALK FORWARD: they straighten up and walk side by side straight toward the camera through the drifting dust. Ryan on the LEFT, Ava on the RIGHT. Both faces become fully clear and sharp in warm backlight: Ryan serious and scanning, Ava calm and confident looking straight into the lens. Helicopter stays behind them with its rotors still spinning.

DIALOGUE (at 6–7 s, short, over the rotor noise):
Ryan (deep calm gravelly voice, raised slightly): "Boots on the ground. Let's move."
Ava (clear husky voice): "Copy that."

CAMERA
Ground-level wide shot from a 3/4 front angle at waist height, 35mm lens. Starts wide to show the full landing, then slowly pushes in as they walk toward camera, ending in a medium shot (waist up) with both faces sharp. Shallow depth of field in the final second, helicopter slightly soft behind them. Light handheld shake from the rotor wash, realistic film grain, modern war-film color grade.

AUDIO
Deep rhythmic rotor thump "whop-whop-whop", high turbine whine, roaring wind, gravel spraying, heavy boot impacts on landing, gear rattling as they walk. No background music.

STRICT
Only these two operators exit; no extra soldiers. Same faces, ages and outfits as the references. Photorealistic. No text, no subtitles, no tail numbers, no lettering, no logos, no watermark.

The result: same faces

Not perfect? In my clip the camera angle jumped in the first second. Don’t waste credits regenerating it. Drag the start of the clip in Flow’s timeline, or cut it in any video editor. One second gone, problem solved.

Bonus: how real does the skin look?

I pushed Flow further with an extreme close-up through a magnifying glass. Pores, freckles, even a tiny pimple stayed sharp, and the face never changed.

Step 7: Can Dreamina Do the Action Scene?

For the firefight I tried Dreamina. It has Elements: you upload each reference and give it a name and a description. Two limits to know:

• The element description takes only 800 characters, so I asked Claude to compress each lock.
• The main prompt takes only 4,000 characters. My first one was too long, so Claude cut it down.

Compressed locks (under 800 characters)

@RYAN_REF (757 characters)

SGT. RYAN COLE, 34-year-old American male Special Forces operator. Square jaw, straight nose, deep-set hazel eyes, thick dark brows, small scar through left eyebrow, short dense dark brown trimmed full beard, light tan weathered skin. Short dark brown hair under olive-tan baseball cap. 6'1", broad shoulders, muscular athletic build, heavy steady walk. Multicam shirt sleeves rolled, multicam pants, coyote knee pads, coyote plate carrier with mag pouches, radio on left shoulder, US flag patches both shoulders, silver dog tags, tan gloves, black digital watch, coyote boots, Glock right thigh holster, black M4 on sling. Voice: deep low gravelly baritone, calm, Midwest American accent, short clipped military lines. Serious, protective, always scanning.

@AVA_REF (771 characters)

SGT. AVA REED, 26-year-old American female Special Forces operator. Oval face, defined cheekbones, green-hazel eyes, straight brows, light freckles on nose and cheeks, natural full lips, small scar on chin, fair lightly tanned skin, no makeup. Dark blonde hair in messy low bun, loose wavy strands framing face. 5'8", fit athletic build, strong shoulders, quick confident stride. Multicam shirt sleeves rolled, multicam pants, coyote knee pads, coyote plate carrier with mag pouches, radio on left shoulder, US flag patches, dog tags, olive headset on ears, tan gloves, black watch, coyote boots, Glock right thigh holster, black M4 on sling. Realistic military fit. Voice: clear medium-pitch slightly husky, calm, West Coast American accent, quick precise. Sharp, alert.

@SHADOW6_REF (627 characters)

"SHADOW 6", US Army MH-60M Black Hawk special operations helicopter. Matte dark grey-black paint, heavily weathered, chipped edges, dust and oil stains. Long refueling probe on right side of nose, FLIR sensor ball under nose, external fuel tank on stub wing, four-blade main rotor, tail rotor on left of tail, twin engines on top, sliding cabin door open with M134 minigun mount, dark-tinted cockpit glass, two pilots in olive flight helmets. Sound: deep rhythmic rotor thump "whop-whop-whop", high turbine whine, roaring wind and gravel spray on landing, louder and deeper as it nears. No tail numbers, no lettering, no logos.

5-second firefight prompt (3,640 characters)

@RYAN_REF @AVA_REF @SHADOW6_REF
5-second hyper-realistic cinematic combat scene. Keep Ryan, Ava and Shadow 6 EXACTLY as in their reference images: same faces, ages, beard, hair, outfits, gear and helicopter. Only these two operators on foot. Continues directly after the landing scene.

LOCATION
Same remote desert forward operating base at golden hour. Flat gravel landing zone with small stones. Left background: tall wooden guard tower, long row of tan HESCO barrier walls with desert camouflage net. Foreground left: stacked olive-green ammo crates used as cover. Far background: dry brown rocky mountain ridge. Sun low on the LEFT above the mountains, golden-orange sky, strong warm backlight, long shadows, floating dust lit by sun rays. Shadow 6 sits on the gravel behind them, rotors still spinning, dust swirling.

ACTION TIMELINE (5 seconds, one continuous shot)
0.0–1.0 s: Ryan and Ava walk forward side by side toward camera, Ryan on the LEFT, Ava on the RIGHT. Suddenly enemy tracer rounds zip in from the distant rocky ridge on the right and kick up small dust puffs in the gravel near their boots. Ryan instantly turns his head right, eyes sharp. Ava's loose hair strands whip as she snaps toward the threat.
1.0–2.0 s: Ryan shouts the warning and both drop fast behind the stacked olive-green ammo crates, Ryan kneeling, Ava taking a low crouch beside him, M4 carbines raised to their shoulders in one smooth trained motion.
2.0–4.0 s: Both return fire toward the distant ridge on the right. Controlled bursts from their black M4 carbines with bright orange muzzle flashes, realistic recoil pushing into their shoulders, brass shell casings ejecting and spinning through the golden backlight, falling onto the gravel. Ryan fires steady three-round bursts, jaw clenched, focused. Ava fires precise bursts, cheek on the stock, calm and steady eyes. Behind them, the door gunner of Shadow 6 opens up with the M134 minigun from the open cabin door, a long bright stream of tracers toward the ridge, smoke drifting from the barrels.
4.0–5.0 s: Distant small dust bursts appear on the far ridge where the rounds land (no people visible, no gore). Ryan lowers his rifle slightly and glances at Ava. Ava gives a sharp nod, still aiming. Thin gun smoke and dust hang in the warm light. Hold on both faces, sharp and clearly recognizable.

DIALOGUE
Ryan (at 1.0 s, deep gravelly Midwest voice, shouting over rotors): "Contact right! Get down!"
Ava (at 4.5 s, clear husky West Coast voice, calm): "Ridge is clear."

CAMERA
Starts at a medium-wide front shot at chest height, 35mm lens. At 1.0 s quick handheld whip-follow as they dive behind the crates. From 2.0 s, low side angle from the left at crate level, both operators in profile firing to the right, the helicopter visible behind them. Light handheld shake with each burst, tiny camera jolt on the minigun fire. Final second: slow push-in to a medium close shot on both faces. Shallow depth of field, cinematic film grain, modern war-film color grade, warm orange highlights, deep shadows.

AUDIO
Sharp crack of incoming rounds and gravel impacts, loud realistic M4 rifle bursts with echo off the mountains, metallic tinkle of brass casings hitting stones, roaring buzz of the M134 minigun, deep rotor thump "whop-whop-whop" and turbine whine underneath, heavy breathing, gear rattle as they drop into cover. No background music.

STRICT
Photorealistic, realistic military weapon handling and trigger discipline, no blood, no gore, no visible enemy bodies, no extra soldiers on foot, no slow motion. No text, no subtitles, no tail numbers, no lettering, no logos, no watermark.

My first 8-second attempt failed (“Something went wrong”), and then Dreamina said I didn’t have enough credits. So I went smaller: one 5-second shot on Seedance 2.0 Fast at 1080p, which used 70 of my 120 free daily credits.

Honest verdict: the faces, gear and helicopter stayed consistent, but it looks a little like a video game. For a real movie look you need Seedance 2.5, and the free credits won’t cover it. You can still hit and try with the credits you get every day.

Best Free Tool for a Consistent AI Character: My Verdict

	Google Flow (free)	Dreamina Seedance 2.0 Fast (free)	Seedance 2.5 (paid)
Face consistency	Excellent	Good	Excellent
Realism	Very realistic	Slightly cartoonish	Cinema quality
Free clips per day	About 2 (more with family sharing)	1 short clip	Not free
Best for	Dialogue and character shots	Quick action tests	Final movie shots

For free, Google Flow gives the best consistent characters right now. If you want true cinema quality, Seedance 2.5 is the one, and you can get it inside Dreamina. I’ll compare paid tools properly in a separate video.

Mistakes to Avoid

Problem	Fix
The face changes inside the ChatGPT sheet	Reply: “Keep the exact same face in all 6 portraits and all 3 body views.”
The sheet layout breaks	Ask for the turnaround first, then add the face grid and detail shots.
Two characters’ faces mix in a scene	“Man on the left matches sheet 1, woman on the right matches sheet 2. Keep faces exact.”
The helicopter is parked instead of landing	“Helicopter must be in the air, 1 meter above ground, wheels not touching.”
Faces blurry at the end of a Flow clip	“Final 2 seconds: close medium shot, faces in perfect focus.”
Jump looks rushed in 8 seconds	Split into two clips: landing + jump, then walk forward.
Video tool blocks the action scene	Add “no blood, no gore, no visible enemy bodies”.

• Call characters by name (Ryan, Ava, Shadow 6), never “the soldier” or “the girl”.
• Keep dialogue to one or two short lines per clip.
• Keep the lock text word for word the same in every clip. That’s also what keeps the voice the same.
• Use the same aspect ratio for images and videos (16:9 with 16:9).

Frequently Asked Questions

How do I make a consistent AI character in every video?

Make a character sheet first, then use its face photo as a reference image and paste the same character lock text in every prompt. Never reword the lock.

Is Google Flow free?

Google Flow gives free daily credits (50 a day in my account in November 2026). An 8-second clip on Omni 1.2 Flash cost 24 credits, so about two clips a day. Members of a Google AI Pro family group can use shared credits.

Is Dreamina good for consistent characters?

Dreamina’s Elements keep faces and gear consistent, but on the free Seedance 2.0 Fast model the result looked slightly cartoonish in my test. Seedance 2.5 gives a much more realistic look but is paid.

What is a character lock?

A fixed block of text describing a character’s age, face, hair, body, outfit and voice. You paste it unchanged into every image and video prompt so the AI rebuilds the same person each time.

Can I make the same helicopter or car in every shot?

Yes. Give the vehicle its own reference sheet and its own lock, exactly like a character, including the sound it makes.

Final Thoughts

A consistent AI character isn’t about one magic prompt. They’re about two things you repeat every time: the same reference picture and the same lock text. Copy the prompts above, build your own soldiers, and tag me in the YouTube comments. I want to see what you make.

Want to see these tools used in a full story? Watch my AI film The Ultimate Mind: Part 1 and Part 2.

Disclosure: all characters, images, videos and voices in this article and video were created with AI tools. Tool names belong to their owners; this post is not sponsored.