fairytale video
Make an ad
AI talking ads: dialogue, voice and lip sync
A creator talking to camera is what makes a UGC ad feel like UGC. It's also where AI ads most often go wrong: lines that are too long for the clip, lips that drift from the words, a voice that doesn't fit the face, or extra words nobody wrote. This tutorial covers how to write, voice and check dialogue in an AI ad.
Three examples
A breathless recommendation mid-run, a calm presenter explaining a robot, a gamer's live reaction. Three very different deliveries, each AI-generated and five seconds long, with a line of 8 to 13 words:
- "Guys, you need to try out fairytale video."
- "So now it's picking grapes by itself while we're just walking along."
- "No way. Did you see that? I actually hit that."
Step 1: write less than you think
People talking naturally say about two words a second, and the action needs room too. In the 13 reels on our home page, every line is 5 to 13 words in five seconds. For a 10-second ad, aim for 8 to 20 words, said in one or two breaths. A shorter line also syncs better: the fewer syllables the model has to match to lip movements, the fewer chances to drift.
Step 2: make it sound spoken
Written copy read aloud sounds like an ad. Spoken lines have:
- Contractions: "it's", "I'm", "they're".
- A little filler: most of the lines in our reels open with one, such as "Okay", "Wait", "Guys" or "Oh". It's how real people start talking to a camera.
- Short sentences: "No way. Did you see that?" beats one long sentence.
- No slogans: nobody says "the ultimate all-in-one solution" to a friend.
Read it aloud with a timer. If you trip over it, the model will too. The UGC ad script template has more on word budgets and structure.
Step 3: keep it true
A generated creator can react to what's happening and say what the product is. They can't vouch for experience they don't have: no "I've used this for months", no results, no "my dermatologist recommended it". Write every line so it would be true coming from someone holding the product for the first time.
Step 4: choose a voice that fits
Match the voice to the person on screen and to the moment: bright and quick for a get-ready clip, calm and clear for a presenter, loud for a reaction. Keep the same voice every time the same character appears, so a series of ads feels like one person. Only use a voice you have permission to use, and never imitate a recognisable voice.
Step 5: direct the delivery
Describe how the line is said, not just what it says:
- Energy: relaxed, excited, breathless, deadpan.
- Pace: unhurried, or quick and chatty.
- Eye line: into the lens while speaking. Looking away mid-sentence weakens the sync.
- Placement: which stage the line falls in, so it doesn't collide with the action.
Don't schedule speech during a bite or a sip. Chewing and talking at once looks wrong in real video and worse in generated video. Put the line after the action.
Step 6: plan it
STAGE 1 (0–3 s)
Initial state: @creator jogging toward the lens on a seafront path,
@product in her hand.
Primary action: she slows to a walk and lifts @product into frame.
Camera: selfie at arm's length, bouncing slightly with her steps.
End state: walking, @product beside her face.
Transition: continuous.
Dialogue: none
STAGE 2 (3–7 s)
Initial state: walking, @product beside her face.
Primary action: she talks to the lens, a little out of breath.
Camera: same angle.
End state: smiling, @product still in frame.
Transition: at 7s, continue into stage 3.
Dialogue: "Okay, quick water break. Look how light this is."
STAGE 3 (7–10 s)
Initial state: smiling, @product beside her face.
Primary action: she holds @product steady toward the lens.
Camera: same selfie angle.
End state: the product clearly in view.
Transition: none, the ad ends at 10s.
Dialogue: none
AUDIO
@creator's voice: bright and breathless. Wind and distant waves.
No music. The line is said once, exactly as written, with no
other speech.That last instruction matters: without it, models sometimes add a word or two at the start or end.
Step 7: check every word
- The line is exactly what you wrote, with nothing added or dropped.
- Lips match the words, especially on "p", "b" and "m" sounds, where the lips close.
- The voice fits the face and stays the same voice throughout.
- Names and numbers are pronounced correctly.
Then add captions, so the line works with the sound off.
Talking ads in fairytale video
In fairytale video, put the approved dialogue in Exact spoken lines, or leave it blank for no dialogue. Choose a reel voice from the library, or record or upload a clip of a voice you have permission to use. For spoken ads, each paid Seedance 2.5 render receives the chosen voice sample as reference audio and is instructed to follow its voice style while saying your approved lines, not the sample's own sentence. This guides the result rather than guaranteeing an exact voice match: check every word, the delivery and lip sync before posting. See 13 AI ad examples to hear the reel voices.