Seedance 2.0 Prompt Guide: Tips, Best Practices & Real Examples

Why Prompts Matter in Seedance 2.0

If you've used other AI video models, you already know the frustration: you write a detailed prompt, hit generate, and get something that vaguely resembles what you described but misses half the details. Seedance 2.0 is different. It has some of the best text prompt adherence of any AI video model available today, which means the quality of your prompt directly determines the quality of your output.

A well-structured prompt is the single biggest factor in whether your video looks cinematic or amateurish. Unlike image generators where you can iterate quickly, each video generation takes time and credits, so getting your prompt right on the first try matters.

There's another critical difference between prompting for images and prompting for video: video prompts need to account for motion, timing, camera work, and audio. A beautiful scene description that would produce a stunning image might generate a flat, static-looking video if you don't specify how things move, how the camera behaves, and what the viewer hears.

This guide covers everything we've learned from building and refining dozens of templates in the Starrd library. These aren't theoretical tips — they're battle-tested techniques that produce consistently better results.

The Core Prompt Formula

Every great Seedance 2.0 prompt follows the same fundamental structure:

Subject + Scene/Environment + Action/Motion + Camera Movement + Lighting + Style/Mood

You don't need all six elements in every prompt, but the more intentionally you cover each one, the better your results will be.

A young woman in a leather jacket walks through a neon-lit Tokyo alleyway at night. Rain puddles reflect pink and blue signs. Steadicam tracking shot following from behind. Volumetric fog, shallow depth of field. Cinematic film grain, 24fps. Generate audio.

Let's break down what makes this work: the subject is specific (young woman, leather jacket), the environment is vivid (neon-lit Tokyo alleyway, rain puddles, reflections), the motion is clear (walks), the camera is defined (steadicam tracking from behind), and the style is cinematic (film grain, shallow DOF). Every element is doing work.

Now let's go deeper on each element.

Subject Description

The number one mistake in subject descriptions is being too vague. "A person standing in a room" gives Seedance almost nothing to work with. You need to describe what the camera would actually see.

Be specific about physical details: "a man in his 30s with short dark hair, sharp jawline, wearing a fitted navy suit with an open collar" gives the model a clear picture to render. Describe what you SEE, not what you know — don't write "a confident businessman," write what confidence looks like: "standing tall with shoulders back, chin slightly raised, direct eye contact."

Key details to include in your subject description:

• Build and posture — tall, athletic, slouched, rigid
• Hair — color, length, style (slicked back, messy, braided)
• Clothing — specific garments, colors, textures (not just "dressed well")
• Expression — furrowed brow, slight smirk, wide-eyed (not "happy" or "sad")
• Distinguishing features — tattoos, scars, glasses, jewelry

If you're using Starrd, the AI handles this for you — it analyzes your uploaded photos and writes a personalized subject description that captures your actual appearance. But if you're writing prompts from scratch, investing time in your subject description pays off enormously.

Scene and Environment

Your environment description sets the entire mood of the video. Think of it as production design — you're telling Seedance what the set looks like.

Set the location precisely. "A city street" could be anything. "A rain-slicked downtown Manhattan street at 2 AM, steam rising from subway grates, yellow taxi cabs blurring past" is a scene you can see in your mind.

Include atmospheric details that a cinematographer would notice:

• Time of day — golden hour, midnight, overcast afternoon
• Weather — light drizzle, heavy snowfall, humid heat haze
• Crowd density — empty and desolate, packed shoulder-to-shoulder
• Surface details — wet cobblestones, dusty desert road, polished marble floor
• Background activity — distant city lights, swaying trees, passing pedestrians

One powerful technique is referencing real visual styles. Seedance understands references like "Blade Runner 2049 color palette" or "Wes Anderson symmetrical framing" or "Roger Deakins natural lighting." These references act as shorthand for complex visual languages that would take paragraphs to describe from scratch.

Action and Motion

This is where video prompting diverges most from image prompting. You need to describe movement as a sequence — what happens first, then what happens next.

Use precise, visual verbs. Instead of "moves across the room," write "pivots on his heel and strides toward the window." Instead of "fights the opponent," write "throws a jab, ducks under the counter-punch, then drives forward with an uppercut."

Think about intensity and speed. "Slowly raises a hand to shield his eyes from the light" creates a completely different feeling than "snaps his hand up, blocking the blinding spotlight."

When describing complex action sequences, break them into clear beats. Each beat should be one distinct movement that the model can render cleanly. Three well-described beats will always look better than six vague ones.

Camera Movements — The Biggest Quality Lever

If you take one thing from this entire guide, let it be this: specifying your camera movement is the single most impactful addition you can make to any prompt. A generic prompt with a well-chosen camera movement will outperform a detailed prompt with no camera direction almost every time.

Here are the camera movements Seedance 2.0 handles best, along with when to use each:

Push In / Dolly In

Moves toward the subject, building tension and drawing the viewer's attention. Perfect for dramatic reveals, emotional close-ups, and moments of realization. "Camera slowly pushes in on his face as the realization hits."

Pull Back / Dolly Out

Moves away from the subject, revealing the full scope of the scene. Great for establishing shots, reveals of scale, and endings. "Camera pulls back to reveal the massive arena behind her."

Pan

Horizontal rotation of the camera, left-to-right or right-to-left. Excellent for surveying a scene, following lateral movement, or transitioning between subjects. "Camera pans right across the crowd to find the main character."

Tilt

Vertical rotation, looking up or down. Use for revealing tall structures, power dynamics (tilt up for authority, tilt down for vulnerability), or dramatic reveals. "Camera tilts up from his boots to his face."

Tracking Shot / Steadicam

Follows alongside the subject as they move. The workhorse of cinematic storytelling. "Steadicam tracks alongside her as she runs through the corridor." This is your go-to for any scene where the subject is in motion.

Orbit

Circles around the subject, creating a dynamic, three-dimensional feel. Works brilliantly for hero moments, reveals, and power poses. "Camera orbits 180 degrees around the fighter as he raises his fists."

Crane Shot

Sweeping vertical movement, high to low or low to high. Creates grandeur and scale. "Crane shot descends from aerial view down to street level." Perfect for establishing shots and dramatic arrivals.

Whip Pan

A rapid horizontal snap between two subjects or scenes. Adds energy and urgency. "Whip pan from the pitcher's mound to the batter's box." Use sparingly — it's an accent, not a foundation.

Dolly Zoom

The famous Hitchcock "vertigo" effect where the camera moves in one direction while the focal length shifts in the opposite. "Dolly zoom on his face as the background stretches away." Incredibly dramatic, best used for moments of shock or disorientation.

Handheld

Intentional shake that creates a raw, documentary, or found-footage feel. "Handheld camera follows the action at ground level." Great for fight scenes, chase sequences, and anything that needs to feel visceral and real.

Lighting — Equally Critical to Camera

Lighting descriptions are tied with camera movements for the biggest impact on video quality. If you can only add ONE element to improve your prompt, add lighting. It transforms flat, default-looking renders into cinematic footage.

Here are the lighting setups that Seedance 2.0 renders best:

Golden hour — warm, soft, directional light that makes everything look beautiful. The go-to for outdoor scenes that need to feel aspirational or romantic. "Warm golden hour sunlight streaming from camera left."

Blue hour — the cool, moody twilight just after sunset. Perfect for contemplative, mysterious, or melancholic scenes. "Cool blue hour twilight, deep shadows, city lights beginning to flicker on."

Neon lighting — the cyberpunk staple. Always specify your colors: "neon pink and cyan light reflecting off wet pavement" is much better than just "neon lights." Great for nightlife, urban, and futuristic scenes.

Volumetric lighting — visible light beams cutting through fog, dust, or haze. This is the cheat code for cinematic footage. "Volumetric light beams cutting through dusty warehouse air" instantly elevates any scene.

Studio three-point lighting — professional, controlled, and clean. Specify key, fill, and rim/back lights for maximum control. "Classic three-point studio lighting, strong key light from camera right, soft fill, bright rim light separating subject from background."

Dramatic side-lighting — high contrast with deep shadows on one side of the face. Film noir territory. "Harsh side-lighting from the left, deep shadows consuming the right half of his face."

Overhead ring lighting — creates a distinctive pattern of light and shadow, commonly associated with boxing rings and fighting arenas. "Overhead ring lights casting sharp downward shadows, sweat catching the light."

Pro tip: adding explicit color temperature direction — "color temperature: warm" or "color temperature: cool" — gives Seedance strong visual guidance and produces more consistent results than relying on the model to infer the mood.

A boxer throws a devastating right hook in slow motion. Overhead ring lights create dramatic shadows across his face. Warm amber spotlights from above, cool blue from the crowd. Volumetric haze catches the light beams. Shallow depth of field on impact. Cinematic 24fps. Generate audio.

Time-Segmented Prompts — The Advanced Technique

This is what separates amateur from professional Seedance 2.0 output. Instead of writing one long paragraph and hoping the model paces it correctly, you break your 12-second video into three 4-second segments, giving the model a clear story arc to follow.

The structure maps naturally to three-act storytelling:

• 0-4s: Setup — establish the scene, introduce the character, set the mood. This is your wide shot, your establishing moment.
• 4-8s: Development — the action builds, tension rises, the camera gets closer. Things start happening.
• 8-12s: Climax — the peak moment, the dramatic payoff, the money shot. This is what the viewer remembers.

Each segment should specify its own camera angle, subject action, and any lighting or mood shifts. Think of it as writing three mini-prompts that flow together into one coherent sequence.

0-4s: Wide establishing shot of a massive boxing arena. Camera slowly pushes in through the roaring crowd toward the illuminated ring. Dramatic overhead spotlights cut through haze.
4-8s: Medium close-up, the fighter raises his gloves, shadowboxing. Camera orbits around him. Sweat glistens under warm amber ring lights. Crowd noise builds.
8-12s: The bell rings. Explosive first exchange of punches. Slow-motion impact shot. Camera whip-pans between fighters. Dramatic slow motion on the knockout blow. Crowd erupts.
Style: Rocky meets Creed cinematography. Cinematic film grain, 24fps, shallow depth of field. Generate audio.

Notice how each segment has a distinct camera setup, a clear subject action, and builds naturally into the next. The style line at the end applies globally to the entire video. This segmented approach gives Seedance a clear roadmap for pacing, resulting in videos that feel intentionally directed rather than randomly assembled.

Audio Direction

Most people write "Generate audio" at the end of their prompt and call it done. That's leaving quality on the table. Seedance 2.0 can generate remarkably good audio when you give it specific direction.

Think about three layers of audio:

1. Environmental sounds — what the location sounds like (crowd murmur, wind, rain, traffic, birds)
2. Action sounds — what the subject's actions sound like (footsteps on gravel, leather gloves impacting, sword clashing)
3. Music/score — the emotional backbone (dramatic orchestral, lo-fi beats, heavy bass, acoustic guitar)

Combine all three for maximum impact:

Environmental: roaring arena crowd, echoing announcer voice. Action: leather boxing gloves impacting flesh, heavy breathing, feet shuffling on canvas. Music: dramatic orchestral score building from tension to triumphant crescendo. Generate audio.

Audio direction dramatically improves the cinematic feel of your videos. Even a few extra words describing the soundscape can make the difference between a video that feels like a silent movie with noise added and one that feels like it was professionally produced.

Style Modifiers That Work

These are the style keywords and phrases that consistently improve Seedance 2.0 output:

Technical camera terms:

• "Cinematic film grain, 24fps" — the universal quality booster
• "Shallow depth of field" — blurred backgrounds, professional look
• "Anamorphic lens" — the wide, slightly distorted look of Hollywood blockbusters
• "35mm film" — organic, warm, textured footage
• "Rack focus" — shifting focus between foreground and background

Color and grading:

• "Warm color grading" or "cool color grading" — sets overall tone
• "Desaturated" — muted, gritty, realistic
• "High contrast" — dramatic, punchy blacks and bright highlights
• "Teal and orange color grade" — the classic blockbuster look

Director and film references:

• Referencing specific directors or films works as powerful style shorthand
• "Zack Snyder visual style" — slow motion, desaturated with color pops, epic scale
• "Roger Deakins lighting" — natural, motivated, painterly
• "Emmanuel Lubezki long take" — fluid, unbroken, immersive
• "Wes Anderson framing" — centered, symmetrical, pastel

These references work because Seedance has been trained on a massive corpus of visual media and understands these aesthetic languages. Use them as seasoning, not the whole recipe — a reference combined with your own specific descriptions produces the best results.

Common Mistakes to Avoid

After generating thousands of videos, these are the pitfalls we see most often:

1. Mixing Camera and Subject Movement

If the camera is orbiting your subject, don't also have them running across the frame. Pick one to be the dominant source of motion. A static subject with dynamic camera work, or a moving subject with a stable tracking shot — either works. Both together creates visual chaos.

2. Using the Word "Fast"

We cannot stress this enough. The word "fast" in a Seedance 2.0 prompt reliably degrades video quality. It seems to push the model toward jerky, low-coherence motion. Replace it with specific descriptions: "explosive burst of speed," "quick pivot," "rapid-fire jabs," or simply describe the action with inherently energetic verbs like "snaps," "whips," "lunges."

3. Overloading the Prompt

More detail is not always better. If you describe fifteen different actions, six camera changes, and four lighting setups in one prompt, the model has to compromise on everything. Start with 2-3 key elements done well, then add more in subsequent iterations. A focused prompt beats an exhaustive one.

4. Forgetting Lighting

This is the single biggest missed opportunity. We've seen mediocre prompts produce stunning results simply by adding "volumetric lighting through dusty air" or "dramatic golden hour side-lighting." If your videos look flat and default, the fix is almost always better lighting direction.

5. Vague Descriptions

"A cool scene with some action" tells the model nothing. Every word in your prompt should paint a picture. If you can't visualize it from your own description, neither can the model.

6. No Audio Direction

Writing "Generate audio" alone is like telling a film composer "make music." Specify the mood, the key sounds, the energy level. Even one sentence of audio direction — "Heavy bass, crowd energy, sneakers squeaking on hardwood" — makes a noticeable difference.

Real Prompts from the Starrd Template Library

These are real prompt styles used in Starrd templates. They've been refined through dozens of iterations to produce consistently cinematic results. Watch the actual output videos alongside each prompt.

Epic fantasy wizard duel in a dark stone cathedral. Subject raises a staff, channeling swirling magical energy that illuminates the chamber. Sweeping crane shot descending from the vaulted ceiling to eye level. Magical particle effects — glowing runes, spiraling light trails, crackling energy arcs. Dark atmospheric lighting with sudden bursts of colorful magic casting dynamic shadows. Volumetric light beams through stained glass windows. Lord of the Rings meets Harry Potter energy. Camera orbits during the spell release. Deep bass rumble building to explosive magical impact. 24fps, shallow depth of field. Generate audio.

High-energy underground rap battle in a dimly lit basement venue. Subject steps up to the mic, crowd pressing in from all sides. Handheld camera feel with quick cuts between performer and crowd reactions. Neon pink, blue, and purple lighting raking across faces. Bass-heavy atmosphere, smoke machine haze catching colored light beams. Crowd hands raised, phone flashlights creating a sea of white points. 8 Mile underground energy — raw, gritty, electric. Subject delivers bars with intensity, gesturing with conviction. Crowd erupts on the punchline. Film grain, slightly desaturated with neon pops. Generate audio.

A sports car drifts around a mountain hairpin turn at golden hour. Rear tires break loose, sending a plume of tire smoke into warm sunlight. Camera tracks alongside at wheel level, then crane shot rises to reveal the winding mountain road stretching into the distance. Golden hour side-lighting paints long shadows across the asphalt. Motion blur on the background, sharp focus on the car. Engine roar echoing off canyon walls, tires screaming against pavement. Cinematic 24fps, anamorphic widescreen, warm color grading. Generate audio.

Subject emerges onto a massive festival main stage at night. Camera starts behind them, looking out at a sea of 50,000 people, phone flashlights and LED wristbands creating waves of light. Dramatic push-in as they raise their arms and the crowd responds. Massive LED screens flanking the stage display abstract visuals. Confetti cannons fire as the bass drops. Volumetric stage lighting — laser beams cutting through haze in green, purple, and white. Low-angle hero shot transitioning to wide aerial perspective. Bass-heavy electronic music, crowd roaring. Cinematic 24fps, shallow depth of field on the performer. Generate audio.

Quick Reference Cheat Sheet

| Category | Recommended Keywords | |---|---| | Camera | Push in, pull back, tracking shot, orbit, crane shot, dolly zoom, handheld, whip pan | | Lighting | Golden hour, volumetric, neon (specify colors), side-lighting, rim light, three-point | | Motion | Slow motion, pivots, lunges, snaps, glances, strides, explosive burst | | Style | Cinematic film grain, 24fps, shallow depth of field, anamorphic, 35mm film | | Color | Warm/cool color grading, teal and orange, desaturated, high contrast | | Audio | Environmental sounds + action sounds + music style + "Generate audio" | | References | Director names, film titles, visual styles (use as seasoning, not the whole prompt) | | Avoid | "Fast," mixing camera and subject motion, vague descriptions, prompt overload |

Start Creating

The best way to learn prompting is to see results. Every template in the Starrd library uses these exact techniques — time-segmented prompts, specific camera movements, detailed lighting, and audio direction — all optimized through dozens of iterations.

You don't have to write prompts from scratch. Upload your photos to any Starrd template and the AI personalizes the prompt for you, applying all the best practices from this guide automatically. It's the fastest way to go from "I want a cinematic video of myself" to a result that actually looks cinematic.

Pick a template, upload your photos, and see what Seedance 2.0 can do with a properly crafted prompt.