Skip to content

AI video model library

Seedance 2.0 Prompts for Product Videos

Nine Seedance 2.0 prompts for product and fashion videos, built on the prompt structure in ByteDance's own Seedance 2.0 guide: numbered reference tags, a named subject you reuse, one camera move per shot, and labelled shots for multi-shot ads.

By Drizzle team

Seedance 2.0 prompt rules, checked against the docs

These rules come from ByteDance's BytePlus ModelArk documentation for Seedance 2.0. They are what make a reference-heavy model like this behave.

  • Reference by number. Inputs are numbered per type in upload order: image 1, image 2, video 1, audio 1. The exact spelling depends on where you run it (see the table).
  • Bind each subject once. Define it with two or three stable features ("Define the woman in @Image1 as Subject 1: curly auburn hair, cream sweater") and reuse that name in every shot.
  • Say what to take from each reference. "Reference the camera movement in @Video1" plus "do not copy the person from @Video1" works better than attaching a clip and hoping.
  • One camera move per shot, in plain film terms: close-up, slow push-in, dolly back.
  • Multi-shot = a storyboard. Write Shot 1, Shot 2, Shot 3, each with framing, action and sound. ByteDance calls precise timestamps ("0–3 seconds") unstable, so label shots instead of timing them.
  • Sound has its own syntax. Dialogue in curly braces ({"Still hot."}), one language per line; sound effects in angle brackets (<lid unscrewing>); music in parentheses ((lo-fi beat, low volume)). Turn audio generation on.
  • Say what you don't want: no subtitles, no added logos, no watermark.
  • Use fewer references than the limit. ByteDance suggests about 1–2 character images, 1 scene image, 1 camera video and 1 audio clip.
How to write Seedance 2.0 reference tags by platform
Where you run itImageVideoAudio
BytePlus ModelArk APIImage 1 or [Image 1]Video 1Audio 1
fal.ai@Image1@Video1@Audio1
Replicate[Image1][Video1][Audio1]
The prompts below use the @Image1 form. Rewrite the tags for your platform. Dreamina and CapCut attach references through their editor.

Seedance 2.0 limits to plan around

Before writing a long prompt, check that your inputs fit. The two limits that catch product teams are the 15-second ceiling (on output and on total reference video or audio) and the block on real human faces in references.

Seedance 2.0 input and output limits
LimitSeedance 2.0Note for product video
Output length4–15 sA 15-second ad is one generation; longer needs editing
Reference imagesUp to 9Under 30 MB each
Reference videosUp to 3, 2–15 s each, 15 s totalUp to 200 MB each, MP4 or MOV
Reference audioUp to 3, 2–15 s each, 15 s totalMust come with an image or video; no audio-only input
Resolution480p, 720p, 1080p, 4KFast and Mini: 480p and 720p only
Aspect ratios21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive9:16 for TikTok Shop and Reels
Real human faces in referencesNot supportedUse account-generated or authorized characters; real model photos get rejected
From the BytePlus ModelArk Seedance 2.0 documentation. fal.ai adds its own cap of 12 reference files in total and accepts smaller reference videos.

Single-shot product prompts

One shot, one camera move, one action. These are the prompts to start with: they are cheaper to iterate on, and they show you quickly whether the model keeps your product intact before you spend on longer sequences.

Product hero orbit from one packshot

A clean opening or closing shot for a product ad, made from the packshot you already have.

Attach: @Image1: the product on a plain background.

Define the ceramic coffee grinder in @Image1 as Product 1: matte sage-green body, walnut lid, brushed steel crank.

Product 1 stands on a pale travertine block in a softly lit studio. The camera makes a slow half orbit from the front-left to the front-right at product height, keeping Product 1 centered and the same size in frame. Warm key light from the left, faint rim light outlining the walnut lid. Product 1 does not move, change shape or change color. No text, no logos added, no subtitles, no watermark.
Duration:
5 s
Aspect ratio:
9:16 or 1:1
Resolution:
720p for drafts, 1080p for the keeper
Audio:
Off, or quiet room tone

Check before you use it: Pause at the half-way point. The back of the product is invented, because @Image1 only shows the front. Add a back view as @Image2 if the far side is visible.

Garment in motion: model walks toward camera

Shows how a dress or wide trousers move, which a still cannot. Strong for fashion TikTok Shop and Reels.

Attach: @Image1: an AI-generated model already wearing the garment (approved still). @Image2: the garment flat-lay for detail.

Define the woman in @Image1 as Subject 1: shoulder-length dark hair, wearing the floral midi dress. The dress print, neckline and hem length must match @Image2.

Subject 1 walks slowly toward the camera along a sunlit stone street, relaxed pace, arms swinging naturally. The light chiffon skirt sways with each step and lifts slightly in a gentle breeze from the right, then settles. The camera dollies backward at the same speed, keeping her framed from head to ankle. Late-afternoon backlight, soft warm tones. No text, no subtitles.
Duration:
5–6 s
Aspect ratio:
9:16
Resolution:
720p
References:
2 images

Check before you use it: Watch the hem. Fabric physics that look right in the first second often lengthen the skirt or merge it with the legs mid-stride.

Unboxing with hands only

A faceless unboxing for products with satisfying packaging. Avoids presenter issues entirely.

Attach: @Image1: the closed box. @Image2: the product out of the box.

Define the box in @Image1 as Package 1 and the product in @Image2 as Product 1.

Top-down shot on a light wooden desk. Two hands lift the lid of Package 1 straight up and set it to the right. The hands fold back a layer of white tissue paper, left side first, then right. They lift Product 1 out with both hands and hold it toward the camera for a beat, label facing up. Soft daylight from a window above. Slow, deliberate movements, nothing falls or slides. <soft paper rustle> <cardboard lid sliding off> No music, no subtitles.
Duration:
8–10 s
Aspect ratio:
9:16
Resolution:
720p
Audio:
On, sound effects only

Check before you use it: Count fingers at the moment the hands grip the product, and check the label against @Image2. Lids that dissolve into tissue paper are common in long single takes; shorten it if that happens.

Texture macro: cream, serum or fabric

A two-to-five-second sensory insert to cut into a longer ad. Works for skincare, food and textiles.

Attach: @Image1: the jar or product, for color and packaging continuity.

Macro close-up. A fingertip presses slowly into the surface of the white face cream from the open jar in @Image1, lifts, and leaves a soft peak that holds its shape. The cream is thick and glossy, with a slight sheen under a single soft light from above. The camera is still. Shallow depth of field, the jar rim blurred in the foreground. <quiet tap of the jar on marble>
Duration:
4 s
Aspect ratio:
1:1 or 9:16
Resolution:
1080p if it is the hero insert
Audio:
Optional

Check before you use it: Texture is a product claim. If the real cream is runny, a stiff peak is misleading. Compare against real footage of the product.

Multi-shot ad prompts

Write multi-shot prompts as a storyboard with numbered shots, each with its own framing, action and sound. Define the presenter and product once at the top and reuse the same names in every shot so they stay consistent.

Three-shot UGC-style product ad with dialogue

A complete short ad in one generation: hook, demonstration, close. Use a fictional presenter.

Attach: @Image1: an AI-generated presenter (not a real person). @Image2: the product.

Define the woman in @Image1 as Subject 1: late twenties, curly auburn hair, cream knit sweater. Define the insulated tumbler in @Image2 as Product 1: matte black, bamboo lid.

Shot 1: medium close-up, handheld phone feel. Subject 1 sits in a car in the driver's seat (parked), holds Product 1 up next to her face and says {"This has been in my car since six a.m."}
Shot 2: close-up of her hands unscrewing the bamboo lid; a little steam rises from the coffee. <lid unscrewing>
Shot 3: medium shot, she takes a sip, smiles, and says {"Still hot."} She tilts Product 1 toward the camera, label facing out.

Natural daylight through the windshield. One camera movement per shot. No subtitles, no logos added, no watermark.
Duration:
12–15 s
Aspect ratio:
9:16
Resolution:
720p
Audio:
On (dialogue + effects)

Check before you use it: The dialogue is a claim. "Still hot" after hours must be true for your product. Also check that Product 1 stays the same color and size across all three shots.

Outfit sequence: three looks, one character

A try-on style video showing a capsule collection on one consistent fictional model.

Attach: @Image1: the AI model. @Image2, @Image3, @Image4: the three outfits.

Define the woman in @Image1 as Subject 1: tall, short black bob, gold hoop earrings. Keep Subject 1's face and hair identical in every shot.

Shot 1: full-length, Subject 1 stands in front of a white bedroom wall wearing the outfit in @Image2, turns a quarter to the left to show the side.
Shot 2: same framing and wall. Subject 1 now wears the outfit in @Image3, steps toward the camera and stops.
Shot 3: same framing and wall. Subject 1 wears the outfit in @Image4, does a slow full turn, arms slightly away from the body.

Static camera at chest height, soft window light from the left. (upbeat lo-fi beat, low volume) No subtitles, no text.
Duration:
12–15 s
Aspect ratio:
9:16
Resolution:
720p
References:
4 images (well under the limit)

Check before you use it: Garments bleed between shots: the second outfit borrows a collar from the first. Check each shot against its own reference, and drop to two looks if the third keeps drifting.

Problem, then product: two-shot demo

A classic before-and-after structure for products that fix a visible problem (lint, tangles, spills).

Attach: @Image1: the product.

Define the fabric shaver in @Image1 as Product 1: white body, clear lint chamber, gray blade guard.

Shot 1: close-up of the sleeve of a navy wool sweater covered in small pills. A hand runs a thumb across them. Flat, slightly cool light.
Shot 2: same sleeve, same framing. Product 1 glides along the sleeve from left to right in one slow pass; behind it the wool is smooth and the pills collect visibly in the clear chamber. Warmer light. <low electric hum>

No text, no subtitles, no music.
Duration:
8 s
Aspect ratio:
9:16 or 1:1
Resolution:
720p
Audio:
On, effects only

Check before you use it: Generated "after" results are demonstrations, not evidence. Keep the result realistic, and don't publish an exaggerated transformation as proof.

Prompts that use video and audio references

Video and audio references are where Seedance 2.0 differs most from image-to-video models. Give each reference one job, state what to take from it, and state what to ignore.

Copy a movement from a reference clip

When a gesture or camera move is easier to show than to describe, such as a trending pose sequence or a specific product reveal.

Attach: @Image1: the AI model or product. @Video1: a 3–8 s clip you own or have rights to, showing the movement.

Define the woman in @Image1 as Subject 1: blonde ponytail, oversized denim jacket, white tee.

Reference the body movement and camera movement in @Video1. Subject 1 performs the same sequence of moves with the same timing: the step forward, the shoulder turn and the hand brushing the jacket open. Keep Subject 1's appearance, clothing and the beige studio background from @Image1. Do not copy the person, clothing or background from @Video1. No subtitles, no watermark.
Duration:
Match the clip (4–15 s)
Aspect ratio:
Match the clip
Resolution:
720p
References:
1 image + 1 video

Check before you use it: The reference person leaking through (their hair, their top) is the main failure. Say explicitly what to take from the video and what not to. Seedance 2.0 rejects reference videos with real human faces, so use footage without a recognizable face or an AI-generated performer.

Cuts timed to a music track

A short product montage where the cuts land on the beat. Useful when you already have licensed music.

Attach: @Image1–@Image3: three product angles. @Audio1: a licensed 10–15 s music clip.

Define the sneaker shown in @Image1, @Image2 and @Image3 as Product 1: white leather upper, gum sole, green heel tab.

Use @Audio1 as the soundtrack. Cut on the strong beats.
Shot 1: Product 1 drops onto a concrete step and lands flat. Close-up, low angle.
Shot 2: side profile, Product 1 rotates slowly on the step.
Shot 3: a foot steps into Product 1 and the laces tighten.
Shot 4: wide shot, someone walks away down a city sidewalk wearing the pair.

Hard light, crisp shadows, slight film grain. No subtitles, no text.
Duration:
10–15 s
Aspect ratio:
9:16
Resolution:
720p
Audio:
@Audio1 drives timing

Check before you use it: Audio must accompany an image or video reference; Seedance 2.0 doesn't take audio alone. Check the logo and heel tab in the fast shots, where details blur the most.

Running these ideas in Drizzle

Drizzle does not offer Seedance 2.0. These prompts are for Seedance on Dreamina, CapCut or an API host. The structure carries over, though: define the product once, lock what must not change, one action per shot. Drizzle's video models are Veo 3.1 (Lite, Fast and Quality, with sound) and Gemini Omni Flash, which takes up to seven reference images. A Veo 3.1 Fast clip costs 22 credits (about $1.10–$1.47).

Compare costs on the Seedance 2.0 pricing guide, see Seedance 2.0 vs Kling 3.0, or make the start frame first with the Nano Banana Pro product prompts.

Sources and review date

Prices and limits change often. Everything on this page was checked on November 9, 2026; confirm the current figure with the provider before you buy.

Frequently asked questions

How do you reference an image in a Seedance 2.0 prompt?

By type and number in upload order. On fal.ai write @Image1, on Replicate [Image1], and on the BytePlus API "Image 1". Then give it a role, for example "Define the bottle in @Image1 as Product 1".

How long can a Seedance 2.0 video be?

Between 4 and 15 seconds per generation. Reference videos and audio are also capped at 15 seconds in total.

How do you make a multi-shot video in Seedance 2.0?

Write a storyboard with labelled shots (Shot 1, Shot 2, Shot 3), each with its own framing, action and sound. Define your subject once at the top. ByteDance notes that exact timestamps are unreliable.

Can I upload a photo of a real model to Seedance 2.0?

Not directly. ByteDance says the 2.0 series doesn't support reference images or videos containing real human faces. Use authorized characters or characters generated in the same account.

Does Drizzle run Seedance 2.0?

No. Drizzle runs Veo 3.1 and Gemini Omni Flash for video. A Veo 3.1 Fast clip costs 22 credits (about $1.10–$1.47).

Keep exploring

Put your next idea in motion.

Bring a product photo. Drizzle helps you make the creative.

Start creating