The Seedance 2.0 Prompt Formula (With 10 Examples)

Ask ten people for their best Seedance 2.0 prompt and you will get ten walls of adjectives. Ask what their prompt formula is and most go quiet. That is the actual gap between random results and repeatable ones: not better words, but a fixed structure you fill in the same way every time.

This is the Seedance 2.0 prompt formula I use for every generation, broken into six slots. Each slot answers one question the model would otherwise have to guess. Fill all six and the model stops guessing. Skip three and you are back to rolling dice with adjectives.

TL;DR

• The formula: subject + action + setting + camera + light and mood + audio. Six slots, one sentence or clause each.
• Order matters less than coverage. A prompt that fills all six slots in any order beats an elegant prompt that fills three.
• The camera slot is the most commonly skipped and the highest-leverage. One movement term plus one speed modifier.
• Audio is a real slot on Seedance 2.0 because generation includes native sound. Describe what the audience hears or accept random ambience.
• Ten worked examples below cover products, characters, landscapes, action, and vertical social. Copy the closest one and swap slots.

The six slots, one question each

Every slot exists because it closes off one dimension the model would otherwise fill with a default. Here is what each slot answers, with the failure mode you get when you skip it.

1. Subject: who or what is this about? One concrete noun phrase with two or three identifying details. "A woman" generates a random woman every time; "a silver-haired woman in a red raincoat" generates the same idea of a woman across attempts. Skip the detail and you lose consistency between generations.

2. Action: what happens during the clip? A clip is time, so give the model an event with a beginning and an end: "picks up the phone, hesitates, answers." Skip this slot and you get the classic AI-video failure: a subject standing around while the background flickers.

3. Setting: where and when? Location plus time of day. The setting drives half the color palette on its own. "A kitchen" is a different video at 7 a.m. than at midnight.

4. Camera: how is this filmed? One movement, one modifier: "slow dolly in", "handheld tracking shot", "static wide shot". This slot separates footage that looks directed from footage that looks like a security camera. If camera language is new to you, the camera movement prompt list gives you 50 ready-made options.

5. Light and mood: what does it feel like? Name the light source and quality: "golden hour backlight", "single neon sign", "overcast flat light". Mood words like "tense" or "serene" work here too, but light descriptions are more reliable because they are physical.

6. Audio: what does the audience hear? Seedance 2.0 generates native audio with the video, so an unspecified soundtrack is a wasted channel. "Rain on glass, distant thunder, no music" is a complete audio direction in nine words. Dialogue works as well: write the line and the character speaks it.

The Seedance 2.0 prompt formula template

[subject with 2-3 details], [action with a start and an end],
[location, time of day], [camera movement + modifier],
[light source and quality], [what the audience hears]

One filled example, slot by slot:

A street food vendor with tattooed forearms, flips noodles in a wok
as flames jump, night market stall in the rain, slow push in at
counter level, wok flames and neon signs as the only light sources,
sizzling oil and crowd chatter, no music

That prompt is six slots and zero filler adjectives. Nothing in it is decorative; every phrase closes a decision.

Ten worked examples across use cases

Each example fills all six slots. Copy the closest one, then swap slots one at a time rather than rewriting from scratch.

1. Product hero. A matte black wireless earbud case, opens as both earbuds rise and rotate, seamless white studio, slow orbit, soft even softbox light, a single clean click and low ambient hum
2. Fashion vertical. A model in an oversized cream coat, turns from the mirror and walks toward the lens, loft apartment at dusk, handheld follow in vertical framing, warm practical lamps, heels on hardwood and fabric rustle
3. Food close-up. A slice of basque cheesecake, a fork breaks through the burnt top and lifts a piece, marble cafe counter in the morning, macro push in, window side-light, the crack of the crust and quiet cafe murmur
4. Character moment. An old fisherman with a white beard, coils rope and looks up at the horizon, wooden pier at dawn, slow dolly in to a medium close-up, cold blue morning light, gulls and rope creaking, no dialogue
5. Dialogue scene. A young chef in a stained apron, looks at the burnt dish and says "we start over", steel restaurant kitchen at night, static medium shot, harsh overhead fluorescents, her line plus the exhaust fan drone
6. Landscape establishing. A stone monastery on a cliff edge, morning fog rolls up and past its walls, coastal mountains at sunrise, aerial rising shot, low golden light through fog, wind and faint distant bells
7. Action chase. A cyclist in a yellow jacket, weaves between cars and jumps a curb, downtown street in the rain, fast tracking shot from a car window, gray midday light with headlight reflections, tire spray and passing horns
8. Ad energy. A basketball player mid-crossover, beats the defender and pulls up for a jumper, outdoor court at golden hour, fast arc around the shot, long shadows and lens flare, sneaker squeaks and a crowd roar on release
9. Calm loop. A ceramic teapot on a low table, steam rises and drifts sideways, tatami room in the afternoon, static close shot, soft paper-filtered daylight, water settling and one distant wind chime
10. Story beat. A girl of about eight, releases a paper boat and chases it along the stream, forest edge after rain, smooth gimbal tracking at her eye level, broken sunlight through wet leaves, stream babble and her laugh

Notice what stays constant across all ten: six slots, concrete nouns, one camera idea, named light, explicit audio. The subject matter is doing the variety; the structure never changes.

Where prompts go wrong, slot by slot

After enough failed generations you learn the failure signatures. Flickering background with a frozen subject: missing action. A different-looking character every attempt: subject slot too thin. Beautiful frame that feels dead: no camera movement. Great visuals with a random orchestral bed underneath: you let the model pick the audio. Wrong mood on a technically correct scene: light slot vague or absent.

The debugging rule that follows: change one slot and regenerate. Rewriting the whole prompt after a miss throws away the five slots that worked. Slot-by-slot iteration is how a mediocre first generation becomes a keeper in two or three attempts instead of ten.

The other habit worth building is stealing structure from working prompts. The Seedance 2.0 prompt gallery is a browsable library of community prompts with their finished videos; read them against the six slots and you will see the formula inside almost every strong result. When you want the slots filled for you, the AI video prompt generator drafts a structured prompt from a plain description.

Frequently asked questions

Does slot order matter in a Seedance 2.0 prompt?

Coverage matters far more than order. Subject-first reads naturally and keeps the model anchored, but a camera-first prompt with all six slots filled beats a subject-first prompt with three. Pick an order you can repeat.

How long should the full prompt be?

The template lands between 30 and 60 words. Below that range a slot is usually missing; above it you are typically stacking adjectives inside one slot, which adds noise faster than control.

Do I need the audio slot on every generation?

If you plan to keep the generated sound, yes. Seedance 2.0 renders audio either way, so leaving the slot empty just means accepting whatever ambience the model invents. Even "no music, room tone only" is a useful instruction.

Can I use this formula for image-to-video?

Yes, with one change: the subject and setting slots are partly defined by your uploaded frame, so spend those words on action and camera instead. The image-to-video workflow rewards motion direction more than description.

What goes in the camera slot if I do not know film terms?

Start with three: push in, pan, and static. Those cover most needs. When you want more range, the 50-prompt camera movement list is organized by use case so you can pick by outcome instead of vocabulary.

Why do my characters change appearance between generations?

Thin subject slots. Two or three fixed identifying details (hair, clothing item, one distinctive feature) give the model something to hold constant. For strict consistency across many clips, attach reference images in reference-to-video instead of relying on text alone.

One formula, every generation

Structure is what makes prompting learnable. Adjectives are taste; slots are craft. Write your next ten generations with the six-slot Seedance 2.0 prompt formula — subject, action, setting, camera, light, audio — and compare the hit rate against your last ten freeform attempts. The difference is usually large enough that you will not go back, and from then on every technique you pick up is just a better way to fill a slot. Start with the template above in text-to-video and keep the formula within reach until it becomes muscle memory.

Further reading

• Seedance 2.0 camera movement prompts: 50 copy-paste shots — ready-made options for the camera slot
• Seedance 2.0 prompts guide — the broader prompting handbook
• 7 Seedance 2.0 prompts that actually work — full-scene examples with commentary