NewlyPrompt guide· April 2026

How to prompt
Seedance 2.0

Seedance does not want pretty-sounding keywords. It wants a shot plan. If you are making app ads, launch teasers, or product demos, this is the prompting system we use inside Natively and the exact skill name to ask for.

Quick answer

The best Seedance prompts follow a simple order: subject and action first, then camera movement, then sound, then transitions or shot labels. Inside Natively, tell the agent to use the video-prompt-builder skill to turn a rough idea into a shot-by-shot Seedance plan. That skill is the exact workflow we use internally for structured video prompts.

Why Seedance prompts fail or work

The biggest mistake people make with Seedance 2.0 is prompting it like an image model. Phrases like cinematic, beautiful lighting, and 4K commercial shot are too static on their own. Seedance is a motion model. It needs to know what happens over time.

fal's own Seedance guide puts it plainly: the model wants cinematic direction, not keyword soup. In practice, that means describing what moves, how the camera behaves while it moves, what should be heard, and where the cut lands.

For app marketing, this is good news. Product demos, UI reveals, launch ads, and mascot-based trailers all naturally break into shots and beats. You do not need a film-school brief. You just need a clear sequence.

The four-part prompt structure

This is the structure we recommend before you touch any special effects language.

1

1. Subject and action

Who or what is on screen, and what exactly are they doing? Start with the verb. A hand swipes through a budgeting app. A mascot drops into frame and points at the CTA.

2

2. Camera movement

Tracking shot, slow push, macro close-up, whip pan, handheld, aerial, rack focus. Seedance responds well to camera language that cinematographers would actually use.

3

3. Sound direction

If audio matters, say what should be heard: soft UI taps, crowd swell, ambient office room tone, bass hit on the logo reveal, light synth bed under the demo.

4

4. Transition or ending beat

For multi-shot prompts, label the cuts. For single-shot ads, specify the ending beat so the clip lands cleanly: logo stamp, CTA card, product close-up, or app icon lockup.

The exact Natively skill to use

Inside Natively, the skill you want is video-prompt-builder. In the repo it lives under the seedance-2 skill package, but the actual skill name exposed to the agent is video-prompt-builder.

If you want the fastest path from rough idea to usable Seedance prompt, say this:

Use the video-prompt-builder skill to turn this into a Seedance 2.0 prompt:

Create a 10-second vertical app ad for a habit tracker.
Target audience: busy professionals.
Mood: clean, optimistic, premium.
Show the app dashboard, streak counter, and a satisfying check-in moment.
End on the logo and App Store CTA.

That is better than asking for "a cool prompt" because the skill is designed to translate creative briefs into a structured output Seedance can actually use.

What the skill produces

The reason the skill is useful is that it does more than rewrite your paragraph. It expands your idea into a planning document with four parts.

Shot-by-shot effects timeline

Every shot gets a timestamp, a main effect or camera behavior, a description of the action, pacing notes, and transition logic into the next shot.

Master effects inventory

A summary of all visual techniques used across the full sequence, so you can see whether the edit is overstuffed or too plain.

Effects density map

A timeline view of where the sequence is dense, medium, or clean. This helps you alternate high-energy moments with simpler, more readable beats.

Energy arc

A top-line view of how the sequence opens, peaks, and resolves. This is what keeps short app ads from feeling like random shots stitched together.

For short-form marketing, that structure matters because good ads are not just beautiful. They have pacing.

Prompt templates you can use immediately

Here are three templates that line up with how Seedance is commonly used inside Marketing Studio.

Single-shot product demo

A hand holds a phone showing a budgeting app dashboard. The thumb taps "Save $50" and the weekly goal ring animates to full. The camera pushes in slowly from a three-quarter angle. Soft UI taps and a subtle warm synth hit play under the action. End on the completed goal ring with the logo appearing in the lower third.

Multi-shot launch ad

Shot 1: Close-up of a phone screen lighting up with a streak notification from a habit tracker app. The camera tilts slightly as the screen wakes. A soft notification chime plays.

Shot 2: Medium shot of the user checking off a task, smiling, and sliding the phone into a jacket pocket. The camera tracks sideways with the movement. Ambient city morning sound underneath.

Shot 3: Macro close-up of the streak counter climbing and a confetti burst inside the app UI. Quick digital push-in timed to a bass hit.

Shot 4: Clean end frame with the app icon, logo, and "Build better habits daily" CTA on a bright background.

Reference-to-video prompt

@Image1 is the app dashboard and should remain the hero UI. @Image2 is the brand color and lighting reference. @Audio1 is the voiceover rhythm guide. Create a 9:16 launch ad where the camera slowly pushes toward the phone, the dashboard widgets animate in sequence, and the final frame lands on the premium analytics screen. Keep the motion clean, premium, and minimal. Use soft UI clicks and light electronic ambience timed to the voiceover pacing.

Common mistakes to avoid

1

Using image-model adjectives as the whole prompt

Pretty, cinematic, detailed, high quality, 4K is not a shot plan. Those words can support a prompt, but they cannot carry it.

2

Stuffing too many actions into one shot

If the product spins, the camera whips, the logo appears, the user smiles, and the background explodes all in one beat, the model has to compromise. Give each shot one primary action.

3

Ignoring sound when the clip needs it

Seedance can generate audio. If sound matters, direct it. Without cues, you usually get something generic.

4

Ending without a payoff

App ads need an end beat. If you do not specify the final frame, the clip may end mid-motion and become hard to use in a real campaign.

And if your idea involves a public figure, borrowed franchise character, or third-party brand world, read the restrictions guide before you burn credits.

A practical app-marketing workflow

This is the version of the workflow that gives most teams the best results.

1

1. Start with a plain-English brief

What is the app, who is it for, what are the 2-3 moments you need to show, and what is the final CTA? That is enough.

2

2. Ask for the video-prompt-builder skill

Let the agent turn that brief into a proper Seedance shot plan instead of manually wordsmithing your first draft.

3

3. Iterate in Fast, finalize in Standard

Generate quickly while you are testing shot logic and pacing. Switch to the higher-quality render only after the prompt is doing the right thing.

4

4. Reuse the same prompt skeleton across variants

Swap the hook, the aspect ratio, or the final CTA, but keep the same structure. This is how you produce ad variants without rewriting from scratch.

If you also need stills for the same campaign, pair this with the Nano Banana 2 image guide. If you want the full model and pricing walkthrough, use the Seedance video guide.

Sources & further reading

Official product pages, APIs, and background reading for models and tools mentioned in this guide. Newly is not affiliated with these vendors; links are for your own research.

Frequently asked questions

What is the best prompt structure for Seedance 2.0?

Start with the subject and action, then describe camera movement, then sound, then shot transitions. Seedance performs better with cinematic direction than with comma-separated image-model keywords.

What skill should I use in Natively for Seedance prompts?

Use the skill named video-prompt-builder. In the codebase it lives under the seedance-2 skill package, but the load name exposed to the agent is video-prompt-builder.

Should I write one long paragraph or label shots?

For anything beyond a single simple shot, label your shots explicitly. Shot 1, Shot 2, and Shot 3 style prompts give Seedance clearer cut points and more reliable pacing.

How detailed should a Seedance prompt be?

Detailed enough to specify what moves, how the camera behaves, what the viewer should hear, and how the sequence ends. For most app ads, 2-4 sentences works for a single shot and 4-8 sentences works for a multi-shot sequence.

Can I use the same prompt for text-to-video and reference-to-video?

The core shot language can stay similar, but reference-to-video prompts should explicitly assign roles to @Image1, @Image2, @Video1, and @Audio1 so the model knows what each asset is doing.

What is the biggest prompt mistake people make?

Treating Seedance like an image generator. Prompts like cinematic, 4K, beautiful lighting are not enough. The model needs action, camera movement, sound direction, and transitions.

Tell Natively to use video-prompt-builder.

Bring a rough brief, get back a Seedance-ready shot plan, then generate the ad inside Marketing Studio.