Skip to main content

Seedance 2.0 Prompt Guide for Ads: 12 Prompts That Actually Convert

A practical prompt anatomy for Seedance 2.0 video ads plus 12 copy-paste prompts for product demos, UGC-style scenes, and b-roll, with the exact credit cost to run each one.

Mauricio Valdivia

Mauricio Valdivia

·Updated ·12 min

Seedance 2.0 Prompt Guide for Ads: 12 Prompts That Actually Convert

Two Prompts, Same Credits, Very Different Ads

Two marketers sit down with the same product photo, the same budget, and the same text box. One types "a woman shows a skincare product, nice lighting, make it look viral" and gets a glossy nothing that could advertise any brand on earth. The other writes a director's brief and gets a clip she can put spend behind the same afternoon. They used the same model, so the whole difference came from the prompt. Seedance 2.0 rewards prompts built in five layers, and once you know them, the 12 copy-paste ad prompts further down, sorted by job (product demo, UGC-style, b-roll), are built so you can swap in your own product. At the end we show exactly what each clip costs to run. You can test the whole set for less than one hour of a videographer's time.

What Seedance Rewards in a Prompt

Before you get to the prompts, it helps to know what you're talking to. Seedance 2.0 is ByteDance's video model, and the way fal, the platform that hosts it, describes it tells you how to write for it. The model page on fal pitches "cinematic output with native audio, multi-shot editing, real-world physics, and director-level camera control." Each phrase in that sentence points to something your prompt should contain. If the model offers camera control, give it a camera instruction. If it renders physics, describe a physical action it can simulate. If it generates audio, say what the scene sounds like.

A director's brief, not a wish

The most common prompting mistake in ad work is writing quality adjectives where the directions should go. Words like "stunning", "high quality", and "viral" give the model nothing to execute. They're wishes about the output, and they don't describe a scene. ByteDance has trained its Seedance models for prompt adherence since Seedance 1.0, and the Seedance 1.0 section of the official model page highlights precise prompt following with stable control over a "rich variety of camera movements". So the whole family responds to the vocabulary of a shot list, words like close-up, top-down, slow push-in, and handheld sway. Write the sentence a director would say to a camera operator, and the model has something to follow.

The five layers of an ad prompt

Every prompt in this guide has the same five layers, and they're what you should check your own prompts against.

LayerWhat to writeExample phrase
Subjectone person or product, one actionhand presses one pump of serum
Camerashot type plus movementhandheld vertical close-up
Motionwhat physically changescream spreads along the cheekbone
Lightingsource and moodsoft morning window light
Audioambience, foley, or a linefingertip tap, quiet room tone

A prompt can be short and still complete. Forty words that cover all five layers will beat two hundred words of atmosphere every time.

Why vague prompts read as stock footage

When a layer is missing, the model doesn't leave it blank. It fills the gap with the most statistically average choice it has, and the average of all advertising is exactly that polished, weightless look people scroll past. Adding words won't fix it. You have to make more decisions, which means picking the shot, the action, and the sound. If you're still deciding whether this model fits your stack at all, our posts on what Seedance 2.5 changes and the Seedance and Veo head-to-head cover the wider model landscape in more depth than we do here.

There's also a quick audit you can run on any draft. Delete every word that fails the question "could a camera operator act on this?" Most drafts trip on the same few words. Here's what to do with each.

  • Cinematic, stunning, premium: these are style wishes with no shot attached. Replace them with a named shot and a named light source.
  • 8k, ultra realistic, high quality: resolution is a render setting and has nothing to do with scene direction, so these words buy you nothing.
  • Viral, scroll-stopping, engaging: these are outcomes, and the model can't target an outcome. Your hook comes from the action you stage in second one.
  • Beautiful person, perfect skin: this is vague casting. Give an age range and one wardrobe note instead.
  • Trending style: the feed decides what's trending, and no prompt can control that.
AI-generated UGC-style creators talking to camera in a row of video cards
Novoads · UGC video ads with AI, ready in minutes.
Try now

The Anatomy, Applied to a Real Ad

Theory is cheap, so let's write the same skincare ad twice. The weak version is the prompt most people type on day one. The strong one shows what the five layers look like when they're all there.

The weak version

"A woman shows a skincare product, nice lighting, make it look viral."

Read it against the table above and count the gaps. There's a subject, but there's no action, no camera, no motion, no lighting source, and no sound. That leaves four of the five layers to the model's averages. The clip you get back will be smooth, pretty, and unusable, because nothing in it belongs to your brand or your angle.

The strong rewrite

Prompt 1 · Skincare hero
Handheld vertical close-up of a woman in her late 20s at a bright bathroom mirror. She lifts a small amber serum bottle into frame, presses one pump onto her fingertips, and pats it along her cheekbone. Subtle handheld sway, like a phone held at arm's length. Soft morning window light with a warm glow from the mirror bulbs. Audio: quiet bathroom room tone, a soft fingertip tap on skin, and she says one short line: this is the only step I never skip.

Read it layer by layer

Map the rewrite back to the anatomy and you'll see every phrase doing a job.

  • Subject: one woman, one bottle, and one action (a single pump, patted in).
  • Camera: a handheld vertical close-up, which is the native framing of a feed ad.
  • Motion: the lift into frame and the patting gesture. Both are physical and small.
  • Lighting: two named sources (the window and the mirror bulbs), so the model doesn't have to guess.
  • Audio: room tone, one foley detail, and one spoken line under ten words.

Nothing here is fancy. Every choice has simply been made, and that's the whole difference between the two prompts. You can apply the same approach mechanically to any product you sell.

The swap map: one skeleton, five categories

Prompt 1 works as a skeleton. You can re-skin it for almost any category by swapping three layers (subject, foley, light) and leaving the camera and structure alone.

  • Beverage: the pump becomes a can cracked open over ice, the foley is the hiss and the pour, and the light moves to bright daylight.
  • Apparel: the serum becomes a jacket shrugged on at a hallway mirror, the foley is fabric rustle, and the light is warm indoor tungsten.
  • Electronics: the pat becomes a case snapped onto a phone, the foley is the magnetic click, and the light is cool desk LED.
  • Home goods: the cheekbone becomes a candle lit with a match, the foley is the strike and the first crackle, and the light dims to evening.
  • Food: the fingertip becomes a fork pulling a slice, the foley is the cheese stretch and a plate clink, and the light is bright and overhead.

The skeleton keeps the craft in place while the swaps put your brand in. That's why the anatomy matters more than any single prompt in this guide. Once you've memorized the five layers, every new product becomes a fill-in exercise instead of a blank page.

Product Demo Prompts

Demos are the workhorse of direct response, and they have one rule. The viewer has to see the product do the thing. That means locked or slow cameras, one visible transformation, and foley that sells the physics. These four prompts cover the classic demo formats.

Countertop demo and unboxing

Prompt 2 · Countertop demo
Static tripod shot of a compact white blender on a sunlit kitchen counter. A hand drops in spinach, banana slices, and almond milk, presses the lid, and the blend swirls into a smooth green vortex. The camera slowly pushes in toward the jar as it blends. Bright natural window light in a clean modern kitchen. Audio: the blender's rising hum, then a satisfying click as it stops.
Prompt 3 · Unboxing
Overhead top-down shot of hands opening a matte sage-green box on a linen tablecloth. The lid lifts, tissue paper folds back, and the hands raise a ceramic candle into soft focus. Slow, deliberate movements. Warm afternoon side light with gentle shadows. Audio: paper rustle, a soft thump as the lid sets down, quiet room tone.

The static camera in the blender prompt is deliberate. When the product moves, the camera shouldn't. Save camera movement for scenes where the subject is still.

The visible transformation

Prompt 4 · Before and after
Medium shot of a scuffed white sneaker on a wooden stool. A hand wipes the toe cap with a small foam sponge in three passes, and the leather turns bright white exactly where the sponge travels. The camera holds perfectly still so the change reads clearly. Even garage daylight. Audio: the squeak of foam on leather, a short satisfied exhale.

A transformation prompt lives or dies on the phrase "exactly where". If you tie the change to the motion that causes it, the physics engine renders cause and effect instead of a crossfade.

The macro texture shot

Prompt 5 · Macro texture
Extreme macro shot of a thick moisturizer swatch on a glass surface. A fingertip drags slowly through the cream, leaving a clean ridge that holds its shape. Shallow depth of field, the texture glistening under soft studio light against a cool white background. Audio: a gentle tactile smear, faint ambient hum.

Texture close-ups are the cheapest credibility you can buy. One five-second macro at 720p costs 23 credits, and it does sensory work that a paragraph of product copy never will.

A UGC creator filming a product review at home
Novoads · UGC video ads with AI, ready in minutes.
Try now

UGC-Style Prompts That Feel Filmed at Home

The paradox of UGC-style advertising is that imperfection is the production value. A clip that looks too good reads as an ad and gets scrolled past instantly. So these prompts ask for flaws on purpose, like front-camera grain, autofocus breathing, mixed lighting, and background noise. If the format itself is new to you, our guide on how to create UGC ads covers the strategy side, and these four prompts cover the execution.

Handheld testimonial energy

Prompt 6 · Couch testimonial
Handheld vertical selfie shot of a man in his 30s on a living room couch, evening lamp light, slightly grainy like a phone front camera. He holds a supplement jar up to the lens, taps the label twice, and says one line: I stopped skipping breakfast because of this. Natural pauses, imperfect framing, a TV murmuring somewhere behind him.
Prompt 7 · Home gym check-in
Vertical mirror selfie video in a home gym. A woman in athletic wear films herself holding a resistance band, then steps back and does two slow banded squats. The phone stays visible in the mirror and the autofocus breathes once. Overhead LED light, slightly harsh. Audio: rubber band tension, sneakers on foam tiles, one short spoken line: day twelve, still not bored.

The details that matter most here are the ones a cinematographer would remove, like the visible phone, the autofocus breath, and the TV murmur. Keep them in the prompt, or the model will politely clean them up.

Routine formats

Prompt 8 · Get ready with me
Vertical vanity shot of a woman doing her morning routine, talking casually toward the camera while she applies tinted sunscreen from a small purple tube. She points the tube at the lens for one beat, then keeps blending. Ring light mixed with window light, a lived-in bedroom in the background. Audio: her voice close to the mic, a soft cap click, birds faint outside the window.

The street interview

Prompt 9 · Vox pop
Street interview framing on a city sidewalk. A young man answers a question toward the camera, then holds up a pair of earbuds in their charging case. Passersby blur behind him, slight camera shake, documentary energy. Overcast daylight. Audio: street ambience, distant traffic, and his line: honestly, I thought they were triple the price.

The spoken lines in all four prompts are under twelve words. Short lines come through generation cleanly and leave room for the scene's own sound, and that's what makes a clip feel recorded rather than rendered.

B-Roll and Lifestyle Prompts for the Mid-Funnel

Not every clip needs a pitch. Retargeting sequences, story interstitials, and landing-page loops run on b-roll, which means one sensory moment, no dialogue, and strong sound. These are also the prompts where Seedance's physics rendering earns its keep, because liquids, steam, and fabric are exactly what stock libraries always get almost right.

The pour shot

Prompt 10 · Iced matcha pour
Slow motion vertical shot of iced matcha being poured over a glass of milk on a marble counter, the green cascading and folding into soft clouds. The camera arcs a few degrees around the glass as it fills. Bright, airy cafe light. Audio: the liquid pour, ice cracking as it shifts, calm ambience.

Movement and wearables

Prompt 11 · Morning runner
Tracking shot alongside a runner on an empty morning road, focus on her wrist as she glances at a smartwatch. The camera drifts from her face to the watch and back. Golden hour backlight, long shadows, a touch of lens flare. Audio: rhythmic footsteps, controlled breathing, one soft watch chime.

The multi-shot mini spot

With Seedance 2.0's multi-shot editing, one generation can contain real scene cuts, so a 15-second clip can become a tiny three-act spot. The trick is to number the shots and repeat whatever they share, so the model holds those elements steady across the cuts.

Prompt 12 · Three-shot coffee spot
A 15 second mini spot in three shots. Shot one: hands unbox a black coffee grinder on a kitchen counter in morning light. Shot two: close-up of beans falling into the hopper and the burr spinning, steam rising from a kettle behind it. Shot three: a full cup slides toward the camera and a hand lifts it with a satisfied nod. Keep the same kitchen, the same light, and the same hands across all three shots. Audio: grinder whir, a pour, a final ceramic clink.

The closing instruction, "keep the same kitchen, the same light, and the same hands", is what keeps the spot consistent. Without it, each shot re-rolls the scene, and you end up with three commercials for three different kitchens.

UGC creators each holding a different product up to the camera
Novoads · UGC video ads with AI, ready in minutes.
Try now

When a Prompt Misses: Fix One Layer at a Time

Expect a meaningful share of your first generations to miss in some way. That's normal. The useful response is surgical, so figure out which layer failed and change only that one. If you rewrite the whole prompt, you throw away the four layers that worked and start the search over from zero.

Warped labels and invented text

Label text is the first thing to melt when you describe a product from scratch. The model doesn't know your logo, so it hallucinates a plausible one, and plausible is fatal in an ad. No amount of extra description will fix a label. Switch to image-to-video and start from a photo of the real product. Seedance 2.0's image-to-video mode offers start and end frame control with motion prompts, so the frame you upload anchors every pixel of the product and your prompt only has to direct the motion around it.

Too much motion, melted physics

If limbs smear or objects pass through each other, your prompt is asking for more simultaneous motion than the scene needs. Ask yourself how many things are moving. An ad that converts usually needs exactly one. Cut the secondary movement (background crowds, gesturing hands, a swinging camera) and let the one action carry the clip.

The first second is the hook

An ad prompt has a job the model doesn't know about, which is surviving the swipe. Front-load the action so the pump is pressed or the pour begins within the first second, and never open on an establishing shot. Feeds punish slow starts. If a beautiful clip only gets going at second two, nobody sees it. The same logic behind written ad hooks applies here, frame for frame.

Before you re-run a miss, check the layers in this order.

  • One subject, one action? Cut everything else that moves.
  • Camera named? If the subject moves, lock the camera.
  • Lighting sourced? Name where the light comes from.
  • Audio described? Silent prompts give you clips that feel dead.
  • Action in second one? Move the payoff to the front.

Running These Prompts in Novoads

Every prompt in this guide runs as-is in Novoads. Open a project, choose video generation with Seedance 2.0, paste the prompt, and pick your duration and aspect ratio. Clips run 4 to 15 seconds, and the audio is generated natively alongside the video. Resolution goes up to 720p, and you can pick 9:16, 1:1, 16:9, or one of three more ratios. The prompt field takes up to 4,000 characters, almost nine times the length of the longest prompt in this guide, so you've got room to add brand detail without hitting a wall.

Duration and ratio by placement

Duration and aspect ratio are prompt decisions too, because they change what the clip has to do.

  • Feed hooks (9:16, 5 seconds): one action with the payoff in second one. This is where you test hook variants cheaply.
  • Story and Reels demos (9:16, 10 to 15 seconds): there's room for a demo beat plus a result, and it's where the three-shot mini spot fits.
  • Landing page loops (16:9 or 1:1, 5 to 10 seconds): dialogue-free b-roll that still works on muted autoplay above the fold.
  • Marketplace listings (1:1, 5 seconds): the macro texture shot or the visible transformation, framed square.

Pick the placement first, then write the prompt. A 5-second hook and a 15-second demo of the same product are two different briefs, and you can't get one by trimming the other.

What a test round costs

At the default 720p, a 5-second clip costs 23 credits. That's about a dollar at the Plus rate. A 15-second clip costs 68 credits, or about $3 at the Plus rate. You pay more per credit on Starter and less on Ultra. Novoads starts at $15/month for Starter, which comes with 200 credits per month. Plus is $49/month for 1,000 credits. If you need more volume, Ultra starts at $129/month for 3,000 credits. You can cancel anytime, and even the entry plan covers a hero prompt and its hook tests on your own product. Our Seedance 2.0 pricing guide has the full per-platform cost breakdown, including API rates outside Novoads. If you're weighing the no-cost routes first, we give the honest answer in is Seedance 2.0 free. And whether Seedance or another engine is the better fit for a given ad job is a separate question, which we cover in the Seedance and Kling comparison.

A sensible first session

Once you have the plan's credits, run your first session as a real test.

  • Pick the one prompt closest to your product's format (demo, testimonial, or b-roll).
  • Swap in your own subject, product, and spoken line, and keep every other layer intact.
  • Run it at 15 seconds, which costs 68 credits at 720p. Then spend more credits on 5-second hook variants, at 23 credits each.
  • Judge it the way a media buyer would. Would you put $50 behind this clip today?
A UGC creator filming a skincare product review on a phone
Novoads · UGC video ads with AI, ready in minutes.
Try now

The Prompt Is the Brief

Strip away the novelty and none of this is new. Agencies have always known that an ad's quality is set before anyone touches a camera, in the brief that decides the shot, the action, the light, and the sound. AI video didn't get rid of that discipline. It squeezed it into a text box, and now you can turn a brief into a clip in minutes for about a dollar per five-second clip at the Plus rate. The marketers winning with Seedance 2.0 didn't get there with secret prompts. They kept writing real briefs after everyone else settled for wishes. Steal the twelve above, swap in your product, and start on Starter at $15/month, which comes with 200 credits per month.

Frequently Asked Questions

What makes a good Seedance 2.0 prompt for an ad?

One subject doing one clear action, an explicit camera instruction, described motion, a lighting choice, and an audio cue. Seedance was built to take direction (its model page promises director-level camera control and native audio), so specific filmmaking language works better than piles of adjectives like cinematic or viral.

How long should a Seedance prompt be?

The Seedance prompt field in Novoads takes up to 4,000 characters. But most ad prompts that work land somewhere between 40 and 120 words. Length only helps when it adds a concrete detail to one of the five layers. If you pad a prompt with mood words, you'll tend to blur the result instead of sharpening it.

Does Seedance 2.0 generate sound from the prompt?

Yes. Seedance 2.0 generates native audio along with the clip, so your prompt can call for ambience, foley, and even a short spoken line. When a prompt describes the sound, like a blender hum or a fingertip tap on a label, the clip comes back feeling far more real than a silent one with music added later.

Should I describe my product or upload a photo of it?

If it has a label, upload the photo. Seedance 2.0 supports image-to-video with start and end frame control plus motion prompts, so the real product image anchors the clip and your prompt only has to direct the motion. Products you describe from scratch tend to drift on label text and proportions.

How much does it cost to run one of these prompts?

On Novoads, Seedance 2.0 renders at 720p by default. A 5-second clip costs 23 credits, about a dollar at the Plus rate. A 15-second clip costs 68 credits, or about $3 at the Plus rate. You pay more per credit on Starter and less on Ultra. Novoads starts at $15/month for Starter, which comes with 200 credits per month. Plus is $49/month for 1,000 credits. If you need more volume, Ultra starts at $129/month for 3,000 credits.

Can one prompt produce an ad with several scenes?

Yes. Seedance 2.0 supports multi-shot editing inside a single generation, so one prompt can describe two or three shots that cut together in the same clip. Describe the shots in order and repeat whatever they share, like the same kitchen and the same hands, so the model keeps those elements consistent.

Key Takeaways

  • Seedance 2.0 responds to filmmaking language rather than adjectives. Its own model page promises director-level camera control, real-world physics, and native audio, so a prompt that reads like a director's brief will beat one that reads like a wish.
  • An ad prompt that converts has five layers: one subject doing one action, an explicit camera instruction, described motion, lighting, and an audio cue. If you leave one out, the model fills the gap with a generic choice.
  • Organize your prompts by the job the ad has to do. Product demos need locked cameras and one visible transformation. UGC-style scenes need handheld imperfection and a spoken line. B-roll needs a single sensory moment.
  • When a clip misses, change one layer instead of rewriting the whole prompt. If a product label comes back warped, switch to image-to-video and start from a real product photo.
  • On Novoads, Seedance 2.0 renders at 720p by default. A 5-second clip costs 23 credits, about a dollar at the Plus rate. A 15-second clip costs 68 credits, or about $3 at the Plus rate. That means you can test all 12 prompts for less than one hour of a videographer's time.
Mauricio Valdivia

Mauricio Valdivia

Founder of Novoads

Mauricio is the founder of Novoads, where he works to democratize video advertising with AI for brands in Latin America.