Idea to episode.

Credits are film stock. Most of them get spent finding out that an idea was never going to work, one generation at a time. Writing it down costs nothing. So before you open a generator, put the idea through five questions. They are written to be able to say no.

1. One sentence. Can you say the scene with a person, a want and a problem in it? “A woman finds a key under her mat that isn’t hers.” If it takes a paragraph, it isn’t a scene yet.
2. The first frame. What is on screen in the first two seconds? If you can’t picture it, the scroll wins before your story starts.
3. The turn. Something has to change between the start and the end: a reveal, a choice, a reversal. If nothing changes, it’s a mood board.
4. Small enough to hold. Can it be made with one or two characters and one or two locations? Every new face is a consistency problem. Every new location is another look to keep steady.
5. Twice. Would you watch it twice? The ending should ask a question or pay one off.

Short drama works when every second has a job. Five beats, timed for 60 seconds. For a 30-second piece, halve each one and cut the second escalation.

Plan eight to fourteen shots for a minute. Most video generators hand back clips of a few seconds, and the usable part of each is shorter than that, so a shot of four to six seconds on screen is a good unit to plan in.

Worked example: “The Spare Key”

Hook

Extreme close-up. A key taped under a doormat. A hand peels the tape back.

Who and want

Nia, late, arms full of groceries, just wants to get inside. She stops. Her own key is already in her hand. This one isn’t hers.

The wall

It opens her door anyway. Inside, a second toothbrush she doesn’t own. Her phone lights up: “Did you find it?” Unknown number.

The turn

Across the hall, her neighbour’s door is open an inch. Someone inside is watching.

The cut

Nia locks her door from the inside. Then she hears the key turn from the outside.

One sentence, one location, two faces, a turn and an ending that asks a question. It passes all five questions in part 01, which is why it is worth generating.

Faces drift between generations. The fix is not a longer prompt. It is deciding the character once, in words and in pictures, and then refusing to improvise.

The identity block

Fill this in once per character. Paste it into every prompt for that character, word for word, in the same order. Same words, same face.

NAME:
AGE RANGE:
SKIN TONE:
FACE: (shape, jaw, brows, one detail people would notice)
HAIR: (style, length, colour)
EYES:
BUILD:
MARKS: (scar, freckles, tattoo, glasses)
WARDROBE, THIS EPISODE:
ONE PROP: 

Filled in

Nia, late 20s, deep brown skin, oval face with high cheekbones and a small gap in her front teeth, shoulder-length box braids with copper ends, dark brown eyes, slim build, small gold hoop in left ear only, oversized olive bomber jacket over a grey hoodie, a ring of keys on a red lanyard.

The rules

1. Reference sheet first. Before any scene, generate the character on a plain background in the same light: front, three-quarter, profile and full body. Pick the best of each and keep them. They are the character now.
2. Stills first, motion second. Make each shot as a still from the reference, then animate the still. Image-to-video usually holds a face better than text-to-video.
3. Lock the wardrobe. One outfit per episode. A jacket that changes between shots is a continuity error unless the story asks for it.
4. Regenerate, don’t patch. If a face drifts, go back to the reference and generate again. Adding words to rescue a bad take usually makes the next one worse.
5. Name the files. CHAR_nia_front_v3.png, SH04_nia_door_take2.mp4. You will be glad at two in the morning.
6. Two faces per scene. Every extra character in a shot is another face that can drift. Keep crowds out of focus or out of frame.

The formula

[SHOT SIZE] of [CHARACTER: identity block] [ACTION: one verb], [LOCATION], [TIME + LIGHT], [LENS], [CAMERA MOVE], [MOOD + GRADE]

An image prompt describes the frame. A video prompt describes what changes in it: the action and the camera move. When you animate a still, don’t describe the whole picture again. Say what moves.

Swap anything in {braces}. Keep the light and the grade the same for every shot in a scene.

Establishing and wide

1. P01

   Wide establishing shot of {location} at blue hour, streetlights just coming on, one lit window, 24mm lens, slow push in, quiet and tense, muted teal with sodium orange highlights

   Copy
2. P02

   Extreme wide shot, {character} small in frame crossing an empty {location}, overcast soft light, 35mm lens, locked-off camera, lonely, desaturated

   Copy
3. P03

   High angle wide shot of {location} from a rooftop, wet street reflecting neon signs, 28mm lens, slow crane down, noir mood, deep blacks

   Copy

Medium

1. P04

   Medium shot of {character} {action} at a kitchen counter, warm practical light from a pendant lamp, 50mm lens, static camera, intimate, soft film grain

   Copy
2. P05

   Medium two-shot, {character A} and {character B} facing each other across a small table, window light from the left, 40mm lens, slow dolly in, tense, natural colour

   Copy
3. P06

   Medium shot from behind, {character} stops in a narrow hallway and turns their head, hard top light, 35mm lens, handheld, uneasy

   Copy

Close-up

1. P07

   Close-up of {character} as they realise something, eyes shift left, cold light from a phone screen, 85mm lens, shallow depth of field, static, blue night grade

   Copy
2. P08

   Extreme close-up of a hand peeling tape off a key under a doormat, morning light, macro lens, slow push in, crisp detail

   Copy
3. P09

   Close-up, {character} exhales and looks down, rim light from a window behind them, 85mm lens, static, quiet, warm shadows

   Copy

Inserts and details

1. P10

   Insert shot of a phone on a dark table lighting up with a message from an unknown number, 50mm lens, static, the screen is the only light

   Copy
2. P11

   Insert, a door handle slowly turning from the other side, hallway light leaking under the door, 50mm lens, static, suspense

   Copy
3. P12

   Overhead insert of a coffee cup with lipstick on the rim beside a set of keys, soft morning light, 35mm lens, still life, warm

   Copy

POV and angles

1. P13

   POV shot climbing a dark stairwell, flashlight beam sweeping the walls, 24mm lens, handheld at breathing pace, horror

   Copy
2. P14

   Low angle shot of {character} standing in a doorway, backlit by hallway light, near silhouette, 24mm lens, static, power

   Copy
3. P15

   Over-the-shoulder shot from behind {character B} toward {character A} who is speaking, 50mm lens, focus on {character A}, static, conversation

   Copy
4. P16

   Dutch angle medium shot of {character} on the phone, neon sign flickering through the window, 35mm lens, slow orbit, paranoid

   Copy

Movement

1. P17

   Tracking shot alongside {character} walking fast down a sidewalk at night, shop light sliding across their face, 35mm lens, smooth track, urgent

   Copy
2. P18

   Slow dolly out from {character}'s face to reveal they are alone in a large empty room, 35mm lens, cold overhead light, isolation

   Copy
3. P19

   Orbit shot around {character} sitting on the edge of a bed, morning light through blinds striping the room, 35mm lens, slow 90 degree arc, reflective

   Copy
4. P20

   Whip pan from {character A} to {character B} standing at the door, 24mm lens, handheld energy, surprise

   Copy
5. P21

   Rack focus from a framed photo in the foreground to {character} in the doorway behind it, 85mm lens, static, reveal

   Copy

Light and mood

1. P22

   Medium close-up of {character} in a car at night, streetlights sweeping across their face, 50mm lens, static from the passenger seat, thoughtful

   Copy
2. P23

   Silhouette of {character} at a window during a thunderstorm, lightning lighting the room in flashes, 35mm lens, static, dread

   Copy
3. P24

   Golden hour medium shot of {character} on a front stoop laughing with a friend, low sun flare, 50mm lens, handheld, warm and nostalgic

   Copy

A camera move is a sentence about how the audience should feel. Pick one per shot and name its speed.

The rules

1. One move per shot. Two moves in one prompt usually gets you neither, or a camera that wanders.
2. Name the speed. Slow, steady or fast. “Slow push in” and “fast push in” are different scenes.
3. Lead with it if it’s ignored. If the model keeps dropping the move, put the move at the start of the prompt and try again.
4. Dolly is not zoom. A zoom changes the lens. A dolly moves the camera, so the background shifts against the subject. Say “dolly” when you want depth to change.
5. Static is a choice. The strongest shot in a scene is often the one where nothing moves except the person.

Grade words: muted, desaturated, teal and orange, lifted blacks, crushed blacks, warm film grain, cold blue, high contrast, pastel, bleach bypass.

Pick one light and one grade per scene, and put the same words in every shot prompt of that scene. Matching light sells continuity more than matching details do.

1. One line per shot. Short lines are easier to voice, easier to lip-sync and easier to follow on a phone.
2. Write the subtext. People rarely say what they mean. “Did you eat?” can mean “I’m worried about you.”
3. Read it out loud. If you run out of breath, the line is too long. Cut it.
4. Direct the voice. Give every line a tone, a pace and a volume: “quiet, tired, slow.”
5. Leave silence. A look after a line does more than a second line.
6. Caption everything. Plenty of people watch with the sound off. The story has to survive it.

1. Assemble ugly. Lay every shot in beat order first, untrimmed. Watch it once for story only.
2. Trim to the moment. Most generated clips have a short stretch in the middle where everything works. Cut the warm-up and the drift.
3. Cut on movement. A cut during a turn of the head or a door swinging hides the seam between two generations.
4. J and L cuts. Start the next shot’s sound before its picture (J), or let a shot’s sound run under the next picture (L). It is the cheapest way to make separate clips feel like one place.
5. Four layers of sound. Room tone, foley (footsteps, keys, doors), dialogue, then score. Room tone under everything removes the dead silence that makes AI video feel like AI.
6. Captions that read. Large, high contrast, two lines at most, in the middle of the frame. Keep text and faces away from the bottom and the right edge, where the app puts its buttons. Check it on your own phone before posting.
7. Export for the phone. Vertical 9:16 at 1080 × 1920.
8. End early. Cut before the audience is ready. That is what makes them watch it again.

Replace {n} with your real number. Never round up.

Drama openings

1. “This key isn’t mine.”
2. Close-up of a text arriving: “Don’t come home tonight.”
3. “Everyone on this block has a secret. Mine has a lock on it.”
4. A door opens before anyone knocks.
5. “I only lied once. It was enough.”
6. A framed photo. One face has been scratched out.
7. Two people under one umbrella. Neither is talking.
8. “Who’s at the door?” “Nobody’s at the door.”
9. A wedding ring left on an empty bus seat.
10. “You weren’t supposed to see this.”

Behind the scenes

1. I made this scene with no camera. Here’s the shot list.
2. The prompt that finally kept her face the same.
3. This took {n} generations. Here are the ones I threw away.
4. Stop writing longer prompts. Do this instead.
5. The cheapest trick in AI film is sound.
6. One camera move per shot. Watch what happens when I break that rule.
7. How I kept one character the same across {n} shots.
8. Before I spend a single credit, I ask five questions.
9. Same prompt, two lenses: 24mm and 85mm.
10. Why your AI video feels like AI: the silence.
11. Steal my shot list.
12. I wrote this whole scene as one sentence first.
13. This is what a 60-second drama looks like on paper.
14. Three edits that make separate clips feel like one scene.
15. Rate this opening frame. Be honest.

One episode a week is enough if every episode feeds the rest of the week. The episode is the show. Everything else is how you made it.

Post the misses. Rejected generations are content, and they are the proof that you are actually making things.

Before you generate

• The idea passes the five questions
• A one-sentence logline
• A beat sheet with timings
• An identity block and a reference sheet for every character
• A shot list with one prompt per shot
• One light and one grade chosen for each scene

Generating

• Stills first, from the reference sheet
• Every face checked against the reference before animating
• A few takes per shot at most, then rewrite the prompt
• Files named by shot number and take

The edit

• Assembled in beat order
• Every shot trimmed to its moment
• J and L cuts across the seams
• Room tone, foley, dialogue, score
• Captions checked on a phone

Posting

• The hook is in the first frame
• One line of context in the caption
• At most one ask
• The breakdown scheduled for two days later

Tools, features and prices change every month, so treat the examples as a starting list, not a recommendation. Test any new tool on one shot from your current episode before you pay for a month of it.