How long should an AI influencer video prompt be?

For a 10 to 15 second clip, about 120 to 250 words. More words help only if they are timed actions, quoted lines or sounds; adjectives about the look belong in the image.

How do I add dialogue?

Write the exact words in quotes after "She says:" inside a timed beat, and describe the voice once at the top. Keep each line short enough to say in the beat's seconds.

Is there a free AI influencer prompt generator?

Yes: our video prompt builder is free with no login. Pick a character, format, camera and vibe, type the action, and copy the structured prompt.

What is a good prompt for an AI influencer video?

Q: How do I describe the camera?

Name a physical camera in the scene and where it is: "the phone she holds at arm's length, eye level, the only viewpoint, no cuts". Avoid film jargon the model may turn into cuts.

A good AI influencer video prompt names the reference image, says where the camera physically is, splits the clip into timed beats with one action and one quoted line each, lists a few sounds, and ends with an avoid list. Around 120 to 250 words is enough for a 10 to 15 second clip; the opening image carries the look, so the prompt can spend its words on motion and dialogue.

The template

This is the skeleton behind every video we have published with our own character. Fill in the angle brackets.

References:
@Image1: <your influencer's master image>. Open on exactly this face and outfit.

Visual style: <one sentence, e.g. warm phone-camera realism, soft window light>

<Who she is in one sentence: age range, look, attitude.>
Voice: <e.g. bright, fast, slightly sarcastic American accent, with matching mouth movement>.

One uninterrupted take through the front camera of the phone she holds at arm's length, eye level, the only viewpoint, no cuts.
Sound: <3 to 5 room sounds>.

0-4s: <hook action>. She says: "<opening line that names the format>"
4-9s: <the main action>. She says: "<line>"
9-15s: <the turn: something goes wrong, a reveal, a change of mind>. She says: "<last line>"

Avoid: No subtitles. No on-screen text or interface. No phone, hand or tripod visible in frame. No change to her face.

A filled-in example

References:
@Image1: Juno's master image. Open on exactly this face and outfit.

Visual style: natural phone-camera realism, morning window light, slightly grainy.

Juno is a woman in her twenties with a copper bob, freckles and an oversized grey hoodie; upbeat and a little chaotic.
Voice: bright, quick American accent, with matching mouth movement.

One uninterrupted take through the front camera of the phone she holds at arm's length in her small kitchen, eye level, the only viewpoint, no cuts.
Sound: kettle hiss, a spoon on ceramic, a fridge hum.

0-4s: She leans into the lens holding a mug. She says: "Rating my own coffee. Live."
4-9s: She takes a long sip and nods slowly. She says: "Honestly? Barista level."
9-15s: She glances at the counter, freezes, and lifts a jar of salt she used instead of sugar. She says: "Ignore everything I just said."

Avoid: No subtitles. No on-screen text or interface. No phone, hand or tripod visible in frame. No change to her face.

Why each part is there

References first. The model holds a face far better from an image than from words. Start every video from the same master image; more in character consistency.
One camera line. "The only viewpoint, no cuts" stops the model inventing angles. In our renders it also kept the clip feeling like a real phone video.
Timestamps. In our Seedance renders, timed beats were followed far more reliably than a paragraph. A 15-second clip holds two or three beats; 30 seconds holds about five.
Quoted dialogue. Exact lines in quotes came out lip-synced in our Seedance 2.5 renders. Describe the voice in every prompt, or it changes from video to video.
A turn at the end. The last beat should change something. It gives viewers a reason to rewatch and comment.
The avoid list. Ask for a phone-camera look and the model may paint a phone case or a fake streaming bar into the frame. Forbid it.

Mistakes that cost us takes

Describing a famous-actor type. Faces that resemble a well-known star were blocked by the video model's copyright filter after the full render, again and again.
Franchise props and names. An iconic franchise hat and spoken franchise names were blocked; generic replacements passed.
Leaving the outfit out. When the prompt does not restate it, the outfit drifts.
Asking for music in a talking clip. Keep "no background music" unless you pass a song as an audio reference.

For more structure see the Seedance prompting field guide, browse finished prompts in the prompt library, or build one by picking options in the free video prompt builder.

Where Cast fits

You want this structure without writing it: type one line and Cast's prompt writer expands it into one dense Seedance prompt with the character tied to its reference image, the camera, action beats sized to the clip and a no-on-screen-text guard.
You want camera and mood as chips: 10 cameras (Auto, Selfie / Live, Close-up, Wide, Handheld, Push-in, Chase, Low Angle, Overhead and Held) and 9 vibes.
You want to see a long hand-written prompt next to the video it made, in the prompt library.

Where it does not:

Word-for-word control of the final prompt: Cast's prompt writer expands your line, so a hand-written prompt is better run on a model provider directly.
Prompts that reference your own photo; Cast has no image upload.
Takes longer than 15 seconds.

Make a video · first video free