Seedance 2.0 Camera Movements: A Complete Prompt Guide

The first time one of my clips actually looked shot instead of generated, I hadn't changed the subject at all. Same character, same lighting, same six seconds. All I'd done was replace "camera moves around" with "slow dolly in, ending on a tight close-up." That one swap was the difference between a flat AI demo and something that looked like it came off a real set.

That's the thing nobody tells you: in AI video, the camera is half the performance. Get the Seedance 2.0 camera movements right and even a simple scene feels intentional. Get them vague—or leave them out—and the model invents a generic drift that screams "AI."

The Seedance 2.0 prompt guide gives you a quick camera table to get started. This is the deeper reference: every common move, exactly when to reach for it, the precise prompt phrasing that triggers it reliably, how to combine moves without breaking the shot, and the mistakes that quietly ruin your camera work. Everything here is qualitative and based on hands-on use—no invented specs.

Why Naming the Camera Matters So Much

Seedance 2.0 didn't learn cinematography from a textbook. It learned it by watching an enormous amount of real footage, where camera moves carry meaning—a push-in builds tension, a pull-out reveals scale, an orbit shows off a subject. When your prompt names a move the model has seen a million times, it reproduces the feeling of that move, not just the mechanics.

When you don't name one, the model fills the gap with an averaged, aimless motion. That's the floaty, hand-held-but-not-on-purpose wobble that makes so much AI video look cheap. So the single highest-leverage habit in this entire guide is simple: name exactly one camera move per clip, in plain film language.

Rule of thumb: If your prompt doesn't say how the camera moves, the model decides for you—and it decides boring. One named move beats ten adjectives.

The Complete Camera Movement Reference

Here's the full vocabulary. Each row is a distinct move, what it communicates, and the phrasing that triggers it most reliably in Seedance 2.0.

The key distinction people miss: a dolly moves the whole camera through space; a zoom only changes the lens. A push-in feels like you're stepping toward the subject and the background shifts naturally. A zoom flattens the scene and feels more clinical. Both are useful—just don't say "zoom" when you mean "dolly in," because you'll get the flat version.

When to Use Each Move (The Decision Part)

A reference table tells you what the moves are. Choosing the right one is a storytelling call. Here's how I actually decide.

• Static / locked-off — when the subject's action is the star and you don't want the camera stealing focus: a face reacting, a product on a surface, a slow dramatic beat. It also lets subtle physics read—weight, breath, a small gesture. "No camera movement" is quietly one of the most cinematic choices you can make.
• Dolly in / push-in — the workhorse for emotion. A slow push toward a face pulls the viewer in to what the character feels. Use it for tension, a realization, or an emotional peak. Keep it slow; a fast push-in reads as aggressive.
• Dolly out / pull-out — the natural closer. Pull-outs reveal context ("she's alone in a huge empty stadium") and give a clip a sense of ending. If you want a shot to feel like the last shot, dolly out.
• Orbit / arc — the showcase move for products, characters, or anything you're presenting rather than narrating. It shows dimension and makes a subject feel important—a natural fit when generating a hero shot from text to video.
• Crane — scale and drama. A crane up at the end feels like a curtain call; a crane down feels like an arrival. Use sparingly—it's punctuation, not a default.
• Handheld follow — energy and "this is real." Suits action, walking-and-talking, and anything meant to feel unpolished on purpose. Ask for "subtle" handheld unless you want chaos.
• Tracking — the move for a subject in motion. A tracking shot that keeps pace sells walking, running, or driving far better than a static frame the subject glides across. Match the speed to the action.
• Whip pan — pure energy and transition. Great for fights, reveals, and fast cuts. It's also forgiving: the motion blur hides imperfection, handy for punchy social clips.
• Tilt — reveals verticality (tilt up a skyscraper, tilt down to the hands) and shifts attention within a shot without cutting.

Rule of thumb: Match the move to the intent—push in for emotion, pull out for reveals, orbit to showcase, track to follow, static to observe. If you can't say why the camera moves, lock it off.

Exact Phrasing: How to Trigger a Move Reliably

Two prompts asking for the same move don't perform the same. Over a lot of generations, these phrasing habits gave me the most consistent camera control.

Name the move in film language. "Dolly in," "orbit," "crane up," "whip pan"—these are terms that appear in real footage descriptions, so the model recognizes them. Vague synonyms like "camera moves closer" are hit-or-miss.

Add speed and direction. "Slow dolly in," "180-degree orbit clockwise," "fast whip pan." Speed changes the entire read of a shot, and direction removes ambiguity.

Give the move an endpoint. "Dolly in, ending on a tight close-up." "Crane up to a bird's-eye view." Telling the model where the move lands keeps it from wandering or reversing halfway through.

Anchor it to the subject. "Camera orbiting the seated woman," "tracking alongside the runner." An anchored move stays coherent; an unanchored one drifts.

Respect the motion. If the subject moves, the camera should acknowledge it—"tracking at walking speed," not "static shot" of a walking person, which produces that gliding-across-frame look.

Here's the shape I use:

[camera move] + [speed] + [subject anchor] + [endpoint], [angle], [shot size]

Applied:

slow dolly in on the old fisherman's face, ending on a tight close-up,
low angle, shallow depth of field

For more on how the camera fits the full subject-action-camera-lighting formula, the step-by-step guide to using Seedance 2.0 walks through the whole workflow.

Combining Camera Moves (Carefully)

The instinct once you get comfortable is to stack moves: "orbit while craning up and pushing in." Resist it. In six seconds, the model can execute roughly one clear move well. Ask for three and it usually mushes them into an incoherent drift—the exact thing we're trying to avoid.

That said, some combinations genuinely work because they're compatible, not competing:

• Tracking + slight push-in — follow a walking subject while easing closer. The two moves share a direction, so they blend.
• Crane down into a dolly in — arrive on a scene, then press toward the subject. Sequential, not simultaneous.
• Orbit + subtle tilt — arc around a subject while slowly tilting up for grandeur.

The rule that keeps these clean: combine moves that share intent or run in sequence, never two moves fighting for the same axis. And if you want a genuine multi-move sequence—establish wide, then cut to a push-in—generate it as separate clips and edit them together. One move per generation stays your safest default.

Rule of thumb: One camera move per clip is the rule; a compatible pair is the exception. If two moves pull in different directions, split them into two generations.

Camera Mistakes That Quietly Ruin Clips

Most bad camera work isn't the model failing—it's the prompt. Here are the repeat offenders and their fixes.

No camera direction at all. You described the scene but never the camera, so the model drifts aimlessly. Fix: name one move, always.

Too many moves at once. Orbit + crane + zoom in one clip. Fix: pick one; sequence the rest across clips.

Confusing zoom with dolly. You asked for "zoom in" and got a flat, lifeless push. Fix: say "dolly in" or "push-in" when you want the camera to move through space.

Camera fighting the subject. A static shot of a running person, so they skate across the frame. Fix: match the move to the motion—track a mover, lock off a talker.

Overcooked speed. "Fast dramatic dolly in" on a quiet emotional beat feels jarring. Fix: default to "slow" for emotion, save "fast" for action.

No endpoint. The move starts, then reverses or stalls. Fix: give it a destination—"ending on a close-up," "rising to a wide."

If you're iterating on any of these, change one camera cue at a time and regenerate—otherwise you'll never learn which fix actually worked. That generate-diagnose-adjust loop is worth building the habit around, and it costs less than you'd think once you check whether the plan you're on fits your volume; see is Seedance 2.0 free for where the free tier lands.

Frequently Asked Questions

How do I control the camera in Seedance 2.0? Name the move directly in your prompt using film language—"slow dolly in," "camera orbiting the subject," "handheld follow." Add a speed, anchor it to the subject, and give it an endpoint. That's the whole trick: the camera behaves when you describe it like a director would, not when you leave it to the model.

What camera movements does Seedance 2.0 support? In practice, all the common ones: static/locked-off, dolly in and out (push and pull), orbit/arc, crane up and down, handheld follow, tracking, whip pan, tilt up and down, and zoom. The reference table above lists the phrasing that triggers each most reliably.

Why does my camera movement look random or aimless? Almost always because the prompt didn't name a specific move, so the model invented a generic drift. Add one explicit move with a direction and endpoint—"slow push-in ending on a close-up"—and the aimlessness disappears.

What's the difference between a dolly and a zoom? A dolly physically moves the camera through space, so the background shifts naturally and the shot feels immersive. A zoom only changes the lens, flattening the scene. When you want that stepping-toward-the-subject feel, ask for "dolly in," not "zoom in."

Can I use more than one camera move in a single clip? You can, but keep it to one clear move—or a compatible pair that shares direction, like tracking with a slight push-in. Stacking moves that fight each other produces mush. For a true multi-shot sequence, generate separate clips and edit them together.

How do I get a cinematic camera feel? Match the move to the intent: push in for emotion, pull out for reveals, orbit to showcase, track to follow, lock off to observe. Pair it with a considered angle and shot size, keep the speed appropriate to the mood, and resist over-moving. Restraint reads as cinematic.

Do camera prompts work the same for image-to-video? The camera part does. When animating an image, don't re-describe the scene—just describe the motion and the camera: "subtle push-in, slight parallax, everything else stays still." The prompt guide covers image-to-video prompting in more depth.

The Bottom Line

Camera movement is the fastest upgrade available to your AI video, and it costs you nothing but a few precise words. Seedance 2.0 already knows what a dolly in, an orbit, or a whip pan is supposed to feel like—your job is simply to ask for one, clearly, and let it do the rest.

So build the habit: name exactly one move per clip, in real film language, with a speed, a subject anchor, and an endpoint. Match the move to the intent, combine only when moves are compatible, and split true sequences across generations. Do that consistently and your clips stop looking generated and start looking directed.

Pick one move from the table, describe a single shot around it, and see the difference for yourself:

Start free → Seedance 2.0 AI Video Generator