Seedance 2.5 UGC Ad Prompts with Lip Sync (Tested)

Add to home screen

Open from your home screen next time

  1. 1Open the browser menu ⋮
  2. 2Choose “Install app” or “Add to Home screen”
  3. 3Tap “Add” or “Install” to confirm

Menu labels vary by browser. If the option is missing, open this page in Safari or Chrome.
Blurbly
0
Pricing

Seedance 2.5 · Video prompts

Seedance 2.5 prompts for UGC-style talking ads

A person holds the product, says one line to a phone camera, then uses it. That is the UGC ad in its simplest form, and Seedance 2.5 can make the whole thing from one prompt: write the line in quotation marks and it generates the voice and the lip movement along with the picture. We tested three of these ads, and the clips below keep their sound so you can judge the delivery yourself.

All three clips are 8 seconds at 480p. Only the bathroom clip starts from a photo of the creator. Seedance’s input filter refused our photoreal start frame of the man in the kitchen, so that clip and the car clip use reference-to-video instead: a plain product photo plus the creator described in words. We checked every take by transcribing the audio against the script and lining up the speech with the mouth movement.

Tested 2026-10-19 via the Novita AI API (Seedance 2.5, doubao-seedance-2-5-260628), the provider Blurbly uses for Seedance. Products and brands are fictional and the creators are AI-generated adults; clips are unedited except for compression, with the generated audio kept.

Seedance 2.5 clip with sound: the creator says her line to camera, then squeezes cream onto her fingertip
Output of prompt 01Skincare testimonial in the bathroom

Pick any prompt below and press Use this prompt; it lands in this box. Swap in your own product details and photo before you generate.

Prompt formula

Every prompt on this page follows the same order. Fill each slot for your product and keep the order.

SlotExampleWhy it matters
Product anchorUse image 1 as the product: a plain matte black insulated water bottle with a bamboo screw lid and no printed text or logo anywhere on it.In reference-to-video, tell Seedance what the image is for and describe it again in words, including what is not on it. Without “no printed text”, it printed a made-up brand name on our plain bottle.
Creator and settingA man in his early thirties with a short beard, wearing a plain grey t-shirt, stands in a sunny home kitchen.Age, one or two visual details and an everyday location. State an adult age every time. Ordinary rooms read as UGC; styled sets read as an advert.
Phone framingFilms himself vertically on his phone … handheld selfie framing, slight handheld shake.This is what makes it look self-shot rather than produced. We asked for the shake explicitly in every prompt.
The line, in quotesHe says, with a slightly amused, deadpan delivery: "Left this in my hot car all day. Still had ice at six. That's the review."Seedance speaks text inside quotation marks. Put the delivery before the quote. 13–17 words took 3.5–6.5 seconds in our clips, leaving room for the demo.
One action after the lineThen he unscrews the bamboo lid and tilts the open bottle toward the camera so we can see clear ice cubes inside.Speech first, demonstration second. “Then” keeps the mouth free while talking and gives the last seconds a product shot.
Audio rulesNatural English speech with accurate lip sync, quiet kitchen ambience, no music, no subtitles.Name the language, ask for room tone instead of music and rule out burned-in captions. None of our three clips had music or on-screen text.

3 tested Seedance 2.5 video prompts

Each prompt is the exact text we sent to Seedance 2.5. The video beside it is the unedited output, with the settings we used.

01

Skincare testimonial in the bathroom

TikTok / Reels 9:16 Spark-style ad · Meta Advantage+ creative test · PDP review clip

9:16 (from start frame) · 480p (480×854) · 8 s · image-to-video · Seedance 2.5 · generated audio kept

Start frame: a woman in her late twenties in a bright bathroom holding a sage green DEWLY moisturizer tube next to her cheek
Input
Output
Prompt 
She looks into the phone camera and talks in a casual, friendly tone, holding the green DEWLY tube next to her cheek. She says: "I've used this every morning for two weeks, and my skin finally doesn't feel tight by lunch." Then she squeezes a small dot of cream onto her fingertip and shows it to the camera. Handheld phone-camera framing with slight natural shake, natural window light. Her face, hair and the tube stay the same as in the first frame. Natural English speech with accurate lip sync, quiet bathroom room tone, no background music, no subtitles.

Why it works

  • The quoted line came back word for word, spoken between about 1.0 and 5.3 seconds, and her mouth moves only in that window.
  • Because the start frame already shows her face and the labelled tube, image-to-video kept both identical for all 8 seconds, including the DEWLY lettering.
  • “Then she squeezes a small dot of cream onto her fingertip” gives the last seconds a texture shot, which is what a skincare viewer wants to see after the claim.

Swap for your product

  • Line: keep it to one specific, believable result with a time frame (“two weeks”, “by lunch”). Avoid medical claims your product cannot support.
  • Product action: “squeezes a small dot of cream” → “pumps the serum onto her palm”, “swipes the balm across the back of her hand”.
  • Room: bathroom for skincare, kitchen for food and drink, car for on-the-go products; match the start frame.

Attempts: 1. This is the first generation.

02

Deadpan product review in the kitchen

TikTok / Reels 9:16 hook test · YouTube Shorts ad · Amazon customer-style video

9:16 · 480p (480×854) · 8 s · reference-to-video (1 product image) · Seedance 2.5 · generated audio kept

Reference image: plain matte black insulated water bottle with a bamboo lid on a white background
Input
Output
Prompt 
Use image 1 as the product: a plain matte black insulated water bottle with a bamboo screw lid and no printed text or logo anywhere on it. A man in his early thirties with a short beard, wearing a plain grey t-shirt, stands in a sunny home kitchen and films himself vertically on his phone. He holds the black bottle up to the camera and says, with a slightly amused, deadpan delivery: "Left this in my hot car all day. Still had ice at six. That's the review." Then he unscrews the bamboo lid and tilts the open bottle toward the camera so we can see clear ice cubes inside. Handheld selfie framing, natural morning light, an ordinary kitchen behind him. The bottle stays plain matte black with the same bamboo lid as image 1, with no text on it. Natural English speech with accurate lip sync, quiet kitchen ambience, no music, no subtitles.

Why it works

  • Three short sentences gave Seedance natural pauses; it delivered them as three beats with a deadpan rhythm, and the transcript matched the script word for word.
  • “No printed text or logo anywhere on it”, said twice, kept the bottle plain. Without it, attempt 2 printed an invented brand name near the base.
  • “Clear ice cubes” fixed the ending: in attempt 2 the ice inside the black bottle came out looking like black stones.

Swap for your product

  • Line: the format is setup, proof, sign-off. “Dropped it down the stairs. Not a dent. That’s the review.”
  • Product description: rewrite the anchor sentence with your colour, material, lid, and what is printed on it (or that nothing is).
  • Delivery: “deadpan” → “excited”, “whispering because the baby is asleep”, “matter-of-fact”. One adjective is enough.

Attempts: 3, failed tries included. This is the one we kept.

03

Car selfie with a hair product

TikTok / Reels 9:16 creator-style ad · Instagram Story ad · Shopify PDP social proof

9:16 · 480p (480×854) · 8 s · reference-to-video (1 product image) · Seedance 2.5 · generated audio kept

Reference image: lavender spray bottle of leave-in hair mist with a white sprayer and the word LULL, on white
Input
Output
Prompt 
Use image 1 as the product: the lavender LULL leave-in hair mist bottle. A woman in her early thirties with long wavy brown hair sits in the driver's seat of a parked car on a sunny afternoon, filming herself vertically with her phone held at arm's length. She holds the lavender bottle up to the camera and says: "This is the only thing that stops my hair frizzing in this humidity." Then she sprays it twice onto the ends of her hair and runs her fingers through. Selfie framing, natural daylight through the car window, slight handheld shake. The bottle keeps the same lavender color, white sprayer and LULL label as image 1. Natural English speech with accurate lip sync, no music, no subtitles.

Why it works

  • The car interior, seat belt and side light came entirely from the text and read as a real self-shot clip; the product came entirely from image 1, label included.
  • “The only thing that stops my hair frizzing in this humidity” names a problem and a condition in one breath, and was spoken word for word in about 3.5 seconds.
  • Listing the bottle’s three identifiers (colour, sprayer, label) in the last sentence kept the same lavender bottle in her hand after she lowered it out of frame and brought it back to spray.

Swap for your product

  • Location: “driver’s seat of a parked car” → “a gym locker room”, “a train seat on the commute”, “a hotel bathroom”, matched to when the product is used.
  • Problem line: “stops my hair frizzing in this humidity” → your product’s one problem, said the way a customer would say it.
  • Action: “sprays it twice onto the ends of her hair” → the single most recognisable way your product is used.

Attempts: 1. This is the first generation.

What went wrong in testing

Seedance refused our first kitchen start frame before generating: “the input image may contain real person”. The man in it was AI-generated, yet the request was rejected; the woman in the bathroom frame passed.
We switched to reference-to-video with only a product photo and described the creator in text. Expect the same filter if you upload a photo of a real creator.
In reference-to-video, Seedance printed an invented brand name near the base of our plain, unlabelled bottle.
Say what is not on the product, in both the anchor sentence and the closing lock: “no printed text or logo anywhere on it”. The next attempt was plain.
The same attempt showed the ice inside the black bottle as dark, stone-like lumps, which undercut the whole “still had ice” claim.
Describe the proof object precisely: “clear ice cubes”. When the payoff of the line is something visible, it needs its own adjectives.
In that attempt the first word also came out wrong: the transcript read “Who left this…” instead of “Left this…”.
Listen to every take or run it through a transcription tool against the script. With the shorter, three-sentence line in attempt 3, the speech matched the script word for word.
In the bathroom clip a streak of cream appears on her cheek where the tube touched her face in the start frame.
Small, and arguably realistic, but avoidable: start from a photo where the product is held a few centimetres from the face.
These were the slowest renders in our tests: 4 to 12 minutes per 8-second clip (246 to 715 seconds).
Write and queue several scripts at once at 480p, pick the best delivery, then rerun only that one at 720p if you need it sharper.

When to use another model

Wan 3.0

For longer talking clips (up to 30 seconds at up to 1080p) at a lower price per clip; its text-to-video can include sound. We tested it on product demos, not talking heads.

Kling 3.0

For silent hands-on product shots to cut between UGC lines, where finger detail matters more than speech.

Seedance 2.5 food and drink prompts

For product-only B-roll in the same model, such as pours and close-ups to intercut with the creator.

Do it in Blurbly without writing prompts

  1. Add a plain product photo with + Product, then pick a preset creator with + Model, or ask for a creator in your text and let Blurbly cast one.

  2. Give the points to cover and the place, not a finished script, for example “she says it stops frizz, filmed in her car”. Blurbly writes the spoken lines and fits them to the length.

  3. Check that the specs card says UGC with a creator, then read the spoken lines in the Shot list out loud. If a line claims more than the product does, ask for softer wording in the box under the plan.

  4. Tap Unlock to render the ad.

Get a video like this from one product photo

Blurbly plans the shots, writes the prompts and picks the model. You see the plan before you pay.

Make a UGC-style ad from your product photo

Frequently asked questions

Can Seedance 2.5 make a talking UGC ad with lip sync?

Yes. Put the line in quotation marks and ask for natural speech with accurate lip sync. All three of our final clips spoke the scripted English line word for word, and in the frames we checked the mouth moved only while the voice was speaking.

Can I use a photo of my own creator as the start frame?

You need the person’s permission, and Seedance may still refuse the image: its input filter rejected one of our two photoreal start frames as possibly showing a real person. Reference-to-video with a product photo and a written description of the creator avoided the problem.

How long can the script be in an 8-second clip?

Our lines were 13 to 17 words and took about 3.5 to 6.5 seconds to say, leaving time for one product action. Seedance 2.5 runs up to 30 seconds per clip for longer scripts; we did not test longer clips on this page.

Is the generated voice good enough to run as an ad?

Our clips had clear speech, light room tone and no music, which is usable for testing hooks. Accents and voice types are not selectable in the prompt beyond describing them, so for a campaign you may still record or license a voiceover and keep the visuals.

Do I need to label AI-generated UGC ads?

Do not present an AI-generated person as a real customer. Ad platforms have their own rules for disclosing realistic AI-generated content, and they change; check the current policy of each platform you run on before launching.

Sources