AI video prompts that film like a story, not a screensaver
The short answer
A good AI video prompt reads like a shot list, not a mood board: who is in the shot, the one thing they do, where the camera is, what we hear, and when it cuts. Describe small physical actions instead of named emotions, keep spoken lines under ten words, and leave out things models mangle, like screens, mirrors, crowds and fine hand work. On StreamingWars you can skip all of it: write the idea and the shots are written for you.
- Lengths
- 0:15 to 2:00
- Shape
- 9:16 with sound
- First short
- Free up to 30 s
- Cost
- 50 credits / 15 s
- Every vote earns
- 0.25 credits
- Watch both, then vote
- +0.25 credits
The structure
Video model makers publish their own prompt guides, and the strong ones agree: separate the story from the camera, give every shot a purpose, and time the cuts. A 15-second clip holds about five shots.
| Part | What to write | Example |
|---|---|---|
| Who | One fixed description per person, repeated in every shot | A woman in her 30s, short black hair, yellow raincoat |
| Action | One physical action, with direction | Holds the phone out toward him, frame left |
| Camera | Shot size and angle | Medium shot, eye level, handheld |
| Place and light | Where, when, what light | Rainy bus shelter at night, sodium street light |
| Sound | What we hear, including silence | Rain on the roof, a bus passing, no music |
| Line | Who speaks, under ten words | He says: “It was ringing. For you.” |
| Timing | When each shot starts | Shot 2 at 00:03 |
Before and after
| Weak prompt | Stronger prompt |
|---|---|
| A cinematic, emotional, 8K scene of a sad woman at a bus stop | Medium shot, rainy bus shelter at night: a woman in a yellow raincoat checks an empty bench, then looks down the road. Her shoulders drop. |
| She searches for her phone | Her phone is not on the bench. She pats both coat pockets, then stops. |
| He is furious | His jaw tightens. He sets the cup down too hard and doesn't look up. |
| Epic chase through a crowded market | Wide shot: a man sprints down an empty alley. Cut: close on his feet splashing through a puddle. Cut: he vaults a low wall, landing out of frame. |
| They talk about the letter | She holds up the envelope. He says: “Where did you find that?” Cut to her, silent. |
Two lessons from our own side-by-side tests. First, writing “she checks the bench” made the camera put the lost phone on the bench; describe absence, never a search. Second, every version over-acted a named emotion, so describe what the body does instead.
Words and subjects that make models fail
- Prestige words: cinematic, masterpiece, 8K, ultra-realistic. They push toward a glossy, plastic look.
- Wide shots of someone waking up or establishing scenery as the opener. The first frame has to hook.
- Food being eaten, liquids poured or spilled, screens and readable text, mirrors.
- More than three people on screen, and crowds of any size.
- Fights, dancing, acrobatics and fiddly hand work like tying or picking.
- Slow motion. Real-time movement looks more real.
- Speech in an extreme close-up, or two lines in one shot.
- The same framing twice in a row. Change the angle on every cut.
Prompts by category
| Category | The engine | Prompt tip |
|---|---|---|
| Comedy | One unusual thing, heightened, then a button | Name the game in the first shot; end on a reaction |
| Horror | Normal, a rule, a failed safety move, one reveal, cut to black | About one line per 15 seconds; let silence work |
| Action | A goal, geography you can follow, a costly choice | Show stunts as before and after across two shots |
| Romance | A wall between two people and a choice that costs | End on a cliffhanger that changes what they know |
| Drama | One moment of truth | Stay on a face; say less |
Or skip the prompt
On StreamingWars you don't write prompts at all. In Simple mode you type the idea in plain words and the storyboard, camera directions, sound and timings are written for you in the format the camera understands. In Advanced you write short shot lines (“He comes in from the left. He grabs the key. Looks into the lens.”), pick an angle for each, and press Expand to turn one line into a full shot list you can edit. Every short keeps its recipe, so you can open How it was made and see exactly what was filmed.
In the ring right now
Full rankings
Questions people ask
What makes a good AI video prompt?
A shot list: who, one action, camera position, place and light, sound, any line of dialogue, and when it cuts. Concrete beats poetic.
How long should an AI video prompt be?
Long enough to describe each shot clearly. For a 15-second clip that's about five short shot descriptions plus the fixed look of each person and place.
Is there an AI video prompt generator?
On StreamingWars, Expand does that job: it turns one line into a list of key shots with angles that you can edit before filming. Simple mode writes the whole plan for you.
Should I use words like cinematic or 8K?
Usually not. They tend to push models toward a glossy, artificial look. Describe light and camera plainly instead.
Sources
Watch it. Vote for it. Make something better.
Two shorts. You pick. Winner climbs. Your votes earn credits to make your own.
By StreamingWars · First published October 18, 2026