Most seedance parkour prompts collapse into generic action-movie filler — a figure running across rooftops with no specified route, no material behavior under their feet, and no reason the camera is positioned where it is. Parkour as a discipline is built entirely on discrete contact points: a hand on a specific ledge, weight transferred through a wall run, momentum caught by a railing at the last possible second. A prompt that only says "parkour chase across rooftops" hands Seedance an empty stage and asks it to invent the choreography from nothing, and what comes back is a runner-shaped blur with no traceable path through the space. The prompts that work name the architecture as a sequence of discrete objects to touch — this wall, this railing, this gap, this tile — and specify what happens to the surface on contact, because that contact detail is what separates parkour from a person merely sprinting.
The five prompts below treat parkour as five different structural problems. One times an entire golden-hour rooftop run against a light clock that runs out at nightfall. One puts the camera inside the runner's own hands and feet, turning the environment into a named sequence of nodes — wall run, cat leap, tic-tac, kong vault — that a POV lens can hold in frame. One makes the environment itself the antagonist, warping and collapsing under the runner in a cyberpunk cutscene where camera rotation does as much storytelling as the legs. One proves that a single sentence, if it names the right surface material, can produce believable parkour physics with no timestamped beats at all. And one turns a moving bullet train into a scrolling obstacle course with a countdown clock the runner doesn't control. Together they show that parkour prompts succeed or fail on the same axis: how precisely the contact points are named.
1. The São Paulo golden-hour rooftop chase — vertical continuity against a light clock
See the full prompt on Shoty →
"runs full speed across the rooftops of São Paulo's dense favela skyline as the sun sets blood orange behind the city"
Why this works: This prompt structures its entire 15 seconds as a race between the runner and a light transition — the sun setting from "blood orange" to "electric nightlife purple" is a hard clock running underneath every beat, and the seven timestamped shots are paced to land the final rooftop arrival exactly as that clock expires. This does double duty: it gives the runner's physical urgency an external, visible countdown, and it gives the cinematography a built-in color arc so the shot doesn't need a separate grading instruction.
The route itself is written as an unbroken chain of named surfaces and consequences — corrugated metal roofing that thunders underfoot, a slide under a water tank into a roll, a vault over a wall, a concrete ledge fifteen stories up where "her foot clips the edge and a piece of concrete falls, but she doesn't stop." That last detail matters more than the spectacle beats around it: a near-failure that doesn't stop the sequence reads as competence under real risk, which is the entire appeal of parkour footage. The prompt's biggest jump — "camera orbits her mid flight" over a building gap — earns its slow-motion treatment because five prior beats have already established her as someone who doesn't hesitate at height.
Takeaway: Anchor a parkour sequence to a light-transition clock (sunset to nightfall, dawn to full light) so the environment itself is visibly running out of time alongside the runner. Chain each rooftop transition to a named surface and its physical consequence (corrugated metal thundering, concrete cracking underfoot) rather than writing "runs across the roof" repeatedly, and let one near-failure that doesn't stop the character do more work than a bigger stunt would.
2. The POV architectural node chain — embodied camera and named move vocabulary
See the full prompt on Shoty →
"A parkour expert in POV perspective, defined by visible arms, hands, forearms, shoe tips, lower knees, grips, hand placements, wall runs, precise landings, slides, landing cushioning, and weight shifts."
Why this works: Most action prompts describe a body from the outside. This one deliberately restricts the frame to what a first-person camera can actually see — hands, forearms, shoe tips, knees — and then builds the entire environment as a graph of named parkour techniques the POV lens can hold in view one at a time: wall run, speed vault, precision jump, cat leap, tic-tac, kong vault. Naming the specific technique rather than writing "jumps over it" matters because each of those terms carries a fixed biomechanical shape (a tic-tac pushes laterally off a vertical surface mid-stride; a kong vault clears an obstacle hands-first with the body horizontal), which gives the model an actual motion to render instead of a vague verb.
The environment description reinforces this: "low walls, railing gaps, short height drops, narrow edges... creating a continuous, dynamic path in all directions" describes the city not as scenery but as a route-finding graph, which is exactly how a parkour athlete reads architecture. The one setback in the sequence — a glass walkway that is "slippery from rain" and "begins to collapse under weight" — is resolved through a named recovery action ("catching a lower railing, then swinging to recover"), which keeps the POV disorientation readable as a controlled correction rather than a loss of control.
Takeaway: For POV parkour, restrict the described frame to what a first-person lens can see (hands, forearms, shoe tips) and name the specific technique at each beat (wall run, tic-tac, kong vault, cat leap) instead of generic jumping and running. Write the environment as a chain of touchable nodes, and if you include a failure moment, pair it with a specific, named recovery action so the POV reads as skill rather than chaos.
3. The collapsing cyberpunk cutscene — camera distortion as the antagonist
See the full prompt on Shoty →
"Hero runs along the side of a tilting skyscraper (wall becomes floor). Camera does continuous rolling rotations and sudden drops following the character’s movement."
Why this works: This prompt has no named villain and no creature — the environment itself is the threat, and the camera is the instrument that sells that threat. "Buildings start tilting and warping" converts a static rooftop chase into a fight against architecture, and the prompt escalates the camera language in lockstep with the collapsing geometry: dolly zooms and orbital spins for the vertigo phase, then "continuous rolling rotations and sudden drops" once gravity itself becomes unreliable. The "wall becomes floor" instruction is doing more work than it looks like — it tells Seedance the runner's orientation to gravity is actively changing mid-shot, which is a harder and more specific direction than "the building is falling apart."
The four-beat structure escalates disorientation deliberately before resolving it: establishing high-speed parkour, then vertigo-inducing spins, then the wall-becomes-floor sequence, then a final leap into the void that "resolves into a controlled landing on a lower rooftop as the upper city collapses behind him." That resolution beat is essential — a sequence built entirely from disorientation needs one stable frame at the end, or the viewer never gets confirmation the character survived on skill rather than luck.
Takeaway: When a parkour scene needs stakes beyond human effort, put the geometry itself in motion — tilting, warping, collapsing — and match the camera's rotation and cut rhythm to the environment's instability rather than only speeding up the runner. End on one explicitly "controlled" landing to resolve the disorientation; without it, a sequence built on escalating camera chaos has no way to signal the character regained control.
4. The single-sentence Japanese rooftop run — material physics as compression
See the full prompt on Shoty →
"He leaps between tiled roofs, each landing sending clay tiles sliding and shattering off the edge."
Why this works: Where the São Paulo and cyberpunk prompts run to multiple timestamped paragraphs, this one is three sentences — and it still produces specific, physically legible parkour because it names one exact surface material and lets that material's known behavior generate the rest of the physics. Clay roof tiles are individually seated, brittle, and loosely fixed; naming them (instead of writing "rooftop") tells the model that every landing has a consequence — tiles sliding, cracking, and falling off the edge — without needing a separate physics instruction. That single material choice is doing the work an entire paragraph of impact detail would otherwise require.
The rest of the sentence compounds this economy: "vaults over a wooden railing, drops two stories into a narrow alley, rolls on wet cobblestones sending water splashing" chains three more named materials (wood, stone, water) each with its own one-word consequence (flare, splash), and the prompt closes with "camera tracks from behind in a single continuous motion" — an explicit no-cuts instruction that matches the unbroken physical chain the sentence just described. A parkour run described as one continuous material chain reads as one continuous take almost by necessity.
Takeaway: Before writing a multi-beat parkour sequence, check whether naming the specific surface material at each transition (clay tile, wet cobblestone, corrugated metal, wooden railing) would already generate the physical consequences you'd otherwise need a separate sentence to describe. A short prompt built from three or four named materials in sequence can outperform a long one built from generic verbs.
5. The bullet-train roof obstacle course — obstacle timing on a clock the runner doesn't control
See the full prompt on Shoty →
"Suddenly the train enters a tunnel with a dangerously low ceiling. The courier drops flat and slides underneath while sparks explode inches overhead."
Why this works: Every other prompt in this set lets the runner set their own pace across static architecture. This one puts the runner on top of a moving bullet train, which converts the entire obstacle course into a scrolling hazard arriving at a fixed velocity the character cannot slow down. That single structural choice is what makes the stakes different in kind, not just degree: a low tunnel ceiling isn't a challenge the courier chooses to attempt, it's a hazard already closing at train speed, so "drops flat and slides underneath while sparks explode inches overhead" reads as reaction rather than stunt selection.
The prompt chains obstacles back-to-back with no recovery breathing room — the low ceiling is immediately followed by "another obstacle blocks the roof," forcing a leap with "no time to stop," and that leap goes wrong: the courier slips off the side and hangs "with one hand above the abyss." The recovery is specific and physical ("using momentum and pure strength, they swing themselves back onto the roof"), which keeps the near-death moment readable as skill under pressure rather than a scripted rescue. Unlike the sunset chase and the cyberpunk cutscene, this prompt refuses a settled final image — it ends with the train "rocketing into another tunnel while the courier clings to the front section," sustaining tension instead of resolving it.
Takeaway: Put a parkour sequence on a moving platform (a train roof, a collapsing bridge, a rising elevator shaft) when you want obstacle timing forced onto a clock the character doesn't control, rather than one they choose to run at their own pace. Chain obstacles with no gap between them and give a near-miss a specific physical recovery rather than a cut away from danger, and consider ending on unresolved motion instead of a landing if you want the tension to carry past the final frame.
What these five parkour prompts have in common
- Name discrete contact points, not abstract motion. "Vaults over the wooden railing" and "leaps between tiled roofs" give Seedance a specific object and a specific consequence; "does parkour across the rooftops" gives it nothing to render precisely.
- Match camera perspective to what you want to sell. Third-person tracking sells scale and route (the São Paulo chase); POV sells the biomechanics of a specific technique (wall run, cat leap) that only reads when hands and feet are in frame.
- Surface material is a compressed physics instruction. Clay tile, wet cobblestone, and corrugated metal each carry a known failure behavior — naming the material does the work a separate physics sentence would otherwise need to do.
- To raise stakes, move the environment instead of only speeding up the runner. A tilting skyscraper or a moving train roof forces obstacle timing the character can't control, which reads as more dangerous than a faster sprint across a static rooftop.
- Every near-miss needs a named, specific recovery action. Catching a railing, sliding underneath sparks, swinging back up by momentum — a recovery that's vague ("barely survives") reads as luck; a recovery that names the physical mechanism reads as skill.
- Decide deliberately whether to resolve the tension or sustain it. A controlled landing at the final beat gives visual closure; ending on unresolved motion (still clinging, still running) can carry more tension past the clip's final frame — both are legitimate closing choices, but pick one on purpose.
For adjacent movement discipline and camera technique, see 5 Seedance Skateboarding Prompts for continuous-stunt choreography and 5 Seedance Tracking Shot Prompts for the follow-camera grammar this set relies on throughout. The action scenes use-case gallery collects more high-speed physical prompts, and How to Write Seedance 2 Prompts covers the general prompt-structuring principles behind all five techniques above.
