Most seedance cycling prompts fail because they describe the bike instead of the ride. "A person cycling through the city" gives the model a subject and a verb but no physics: a bicycle only looks correct on screen when the prompt accounts for lean angle through turns, the steady cadence of pedaling, and a surface the wheels can actually grip — asphalt, dirt, a stair rail. Skip those constraints and Seedance defaults to a generic treadmill glide, a rider drifting forward with no sense of weight transferring through the frame.
The five prompts below treat the bicycle as a physics problem with a different solution each time. One keeps a casual city ride deliberately plain so camera smoothness does the work. One turns a bike into an ambient sound event rather than a visual spectacle. One choreographs an entire stunt route through a city as a sequence of named, discrete moves. One builds a downhill descent as a four-beat tracking shot with a mid-air payoff. And one times a delivery sprint to the second, using obstacles as a rhythm track. Together they show that "cycling" is not one register — it ranges from documentary stillness to action choreography to comic timing, and each register needs its own physical vocabulary.
1. The city commute — minimalism as the camera's only job
See the full prompt on Shoty →
"A young East Asian woman riding a bicycle through quiet city streets, casually exploring the neighborhood, cinematic, natural daylight, realistic, smooth camera movement."
Why this works: This prompt is one sentence, and that restraint is the technique. It names a subject, an action, a location, a lighting condition, and exactly one camera instruction — "smooth camera movement" — and stops. No stunt, no plot, no secondary event competing for the model's attention. When a prompt this short still reads as intentional rather than empty, it's because every word it kept is load-bearing: "quiet city streets" sets a traffic-free environment where a bicycle can occupy the full width of the frame without negotiating cars, and "casually exploring" sets a pedaling cadence — unhurried, no urgency in the legs — that a word like "riding" alone would leave ambiguous.
"Natural daylight" and "realistic" do more work than they look like they do. Together they ban the two things that most often break a casual bike shot: an artificial color grade that makes the scene feel staged, and any stylization that would fight against "smooth camera movement" reading as actual camera physics rather than an edited effect. The prompt is betting that if the lighting is unremarkable and the movement is steady, the viewer's attention goes to the only thing left to look at — the rider's posture and the bike's motion through the turn — which is exactly where a commute-register cycling shot wants it.
The absence of a plot is itself a design decision: a commute shot that adds a narrative beat (a near-miss, a destination reveal) stops being a commute shot and becomes an action or vlog clip instead.
Takeaway: For a casual, documentary-feeling cycling shot, resist the urge to add a stunt or story beat — name the environment's traffic density, the rider's pedaling cadence in a word or two ("casually," "unhurried"), and exactly one camera instruction, then stop. Fewer constraints, chosen precisely, keep the register calm; every added beat pulls the shot toward action or narrative instead.
2. The BMX night rider — sound as the only visual confirmation
See the full prompt on Shoty →
"There is a Japanese girl who likes riding her BMX bike at night; if you hear a sound in the quiet of the night, she might be passing by."
Why this works: This prompt never describes what the rider looks like, what she's wearing, or how the camera is positioned — it describes an absence instead, built entirely around the idea that the primary evidence of her passing is auditory, not visual. "If you hear a sound in the quiet of the night, she might be passing by" reframes the entire clip as a listening exercise: the viewer isn't shown a BMX stunt, they're given a quiet nighttime scene and told that a sound is the signal something has happened. This is the opposite structure of the delivery-courier prompt later in this list, which narrates every beat; here, withholding narration is the point.
The phrase "quiet of the night" does the heavy lifting for the visual register before any bike enters the frame — it implies empty streets, minimal ambient light, and no other competing motion, so that a single rider's pass-through reads as the scene's one event rather than one of several things happening. BMX riding specifically (rather than a road bike) matters here too: BMX bikes are associated with tricks, ramps, and curb-hopping, so naming the bike type primes the model toward a more agile, stunt-capable riding style even though the prompt never names a specific trick.
The conditional "might be" is doing something subtle: it keeps the rider's presence uncertain rather than guaranteed, which fits a nighttime mood built on suggestion rather than confirmation. A prompt that said "a girl rides her BMX bike past the camera at night" would guarantee the shot; "might be passing by" leaves room for the clip to feel like an atmosphere piece that happens to include a cyclist, rather than a cyclist shot that happens to occur at night.
Takeaway: When a nighttime or ambient cycling shot needs mood over spectacle, describe the quiet environment and let a sound cue stand in for the visual event — naming the bike type (BMX vs. road vs. fixed-gear) still primes riding style even without describing a specific trick. Conditional language ("might be," "if you hear") keeps a shot feeling like atmosphere rather than a guaranteed narrative beat.
3. The fixed-gear urban descent — a named move for every shot
See the full prompt on Shoty →
"Riding a racing fixed gear bike: lightweight hybrid-material frame, narrow handlebar setup, dark high-reflective metal components."
Why this works: This prompt treats a cycling stunt sequence the way a storyboard treats an action scene: five numbered shots, each assigned exactly one named move — a front-wheel balance glide, a bike rotation midair, a rear-wheel-only glide along a narrow edge, a locked rear-wheel skid past an oncoming car, a final stopped pose with the rear wheel still spinning. Naming the move before describing its context (the stair set, the alleyway, the gap jump) tells the model what the rider's body is doing mechanically before it has to figure out where that body is in space. Without the named move, "rides down a long stairway fast" collapses a dozen possible bike-handling actions into one generic descent.
The "SUBJECTS / ENVIRONMENT / STYLE" header structure separates concerns that would otherwise compete inside one paragraph. The environment block stacks terrain types — stairs, alleyways, underpasses, rooftop — as a continuous route, so the five shots read as one ride through a coherent city rather than five unrelated clips. The style block rules out "realistic 3D rendering" in favor of stylized illustration, which matters because a photorealistic render of shot 4's near-miss with a sedan would read as genuinely dangerous footage; illustration keeps the same beat legible as choreographed action instead.
Shot 4's structure is worth isolating: a car bursts into the frame, the prompt states plainly that a collision seems unavoidable, and then the rider's skid resolves it at the last possible instant. Naming the audience's expected outcome ("the camera, audience, and driver all believe a catastrophic collision is unavoidable") before the prompt contradicts it is a tension-and-release structure borrowed directly from stunt filmmaking — it only works because the setup commits fully to the false outcome before undercutting it.
Takeaway: For multi-shot cycling stunt sequences, assign each shot exactly one named bike-handling move rather than a general description of speed or difficulty, and route the shots through a single continuous environment so the sequence reads as one ride. If a shot includes a near-miss, state the expected bad outcome explicitly before the prompt resolves it — the tension only lands if the setup commits to the danger first.
4. The downhill mountain bike — four beats, one mid-air payoff
See the full prompt on Shoty →
"A dynamic 15-second tracking shot follows a skilled female mountain biker racing down a steep, treacherous dirt trail through a dense pine forest."
Why this works: This prompt divides its 15 seconds into four explicit timestamped beats — 0–4s establishing the descent and tire spray through turns, 4–8s a low side angle on bike control, 8–12s a jump into a backflip silhouetted against sun flare, 12–15s landing and continuing — and the structure matters more than any single beat's description. Beats 1 and 2 are both "riding down the trail," but they change camera position (close chase, then low-and-side) rather than changing the action, which keeps the middle of the clip visually fresh without introducing a new event before the trail has earned one. The jump is reserved entirely for beat 3, roughly two-thirds of the way through — late enough that the descent has established real speed and danger first.
The backflip beat is specified with a lighting condition baked into the trick itself: "silhouetted against a bright sun flare breaking through the forest canopy." This is doing double duty — it's the visual climax of the clip, and it solves the hardest part of filming a backflip, which is making the body's rotation legible against a busy background. A silhouette against a flare simplifies the background to pure light, so the rotating body reads clearly as a shape regardless of what the pine forest behind it is doing.
Beat 4's "lands smoothly, kicking up dust, and continues her breakneck descent" refuses to end on the trick. Ending on the landing, with the ride continuing past it, tells the model the backflip was one obstacle within an ongoing descent, not the destination of the whole clip — keeping the shot reading as "racing downhill," not "a trick video with a bike in it."
Takeaway: For action cycling shots longer than a few seconds, split the duration into beats that vary camera position even while the base action repeats, and place the single biggest trick roughly two-thirds through the clip rather than at the start or the very end. If the trick needs to read against a busy background, specify a lighting condition (backlight, flare, silhouette) that simplifies that background, and let the ride continue past the landing so the trick reads as an obstacle, not the whole point.
5. The delivery courier sprint — obstacles as a timing track
See the full prompt on Shoty →
"A determined mail delivery guy wearing a bright delivery jacket, helmet, backpack, and messenger bag rides a bicycle into frame, checking a small package strapped to the front basket."
Why this works: This prompt breaks a 15-second ride into five three-second blocks, and each block pairs one obstacle with one line of dialogue and one named sound effect — a near-collision with stopped traffic paired with "No way I'm late today," a blocked road and side-lane dodge paired with "Coming through, important delivery!," a curb jump paired with a crowd gasp, and a final brake-and-arrival paired with "Special delivery, right on time." The obstacles aren't random hazards; they're spaced evenly across the runtime like a metronome, so the clip has a readable rhythm — hazard, dodge, hazard, dodge — rather than one long chase with no internal structure.
Pairing dialogue to specific obstacles, rather than letting the rider talk continuously, gives the model natural cut points: each line marks the moment one obstacle resolves and the next begins, segmenting a continuous action sequence into discrete beats without separate shot numbers. The named SFX at every beat — bike bell, tire skid, horn beep, chain spin — work as a cue track too; each sound is tied to a specific physical event rather than generic "upbeat music," keeping every beat grounded in something the bike is actually doing.
The curb-jump beat in block 4 is the clip's one spectacle moment, and it's deliberately smaller in scale than the fixed-gear or mountain-bike stunts elsewhere in this list — "catching air over a cracked sidewalk section" rather than a stairway descent or a backflip. That restraint fits the register: a delivery courier is a comic-action character, not an extreme-sports athlete, and a stunt sized to "slightly late for a delivery" rather than "life-threatening" keeps the tone playful, which the closing victory line and cheerful chime confirm.
Takeaway: For comic or narrative cycling action, divide the clip into even timestamped blocks and give each one exactly one obstacle, one line of dialogue, and one named sound effect — the dialogue and SFX double as segmentation cues, so the model doesn't need explicit shot numbers to read discrete beats. Size the spectacle moment to match the character's stakes; a courier's trick should read as "nearly late," not "extreme sports," to keep the tone consistent.
What these five cycling prompts have in common
- A bicycle only reads as physically real when lean, cadence, and surface grip are named — a plain "riding a bicycle" leaves the model to guess at all three, which produces the generic floating-forward motion that breaks the illusion fastest.
- Restraint is a valid technique, not a missing feature. The commute and BMX-night prompts work precisely because they refuse to add a stunt or a plot; adding either would push the shot into a different register entirely.
- Named moves beat general descriptions of speed or difficulty. "Front-wheel balance glide" and "rear-wheel-only edge glide" tell the model a specific mechanical action; "rides very fast and skillfully" does not.
- Multi-beat action needs a rhythm, not just more events. Evenly spaced obstacles, each paired with one line of dialogue or one camera reposition, give a long action sequence internal structure that a single continuous description can't provide.
- The single biggest trick belongs roughly two-thirds through the clip, after the ride's danger or speed has been established and before the clip has to resolve — too early and it isn't earned, too late and there's no room to land it cleanly.
- Scale the stunt to the character. A fixed-gear stunt rider and a delivery courier can both jump a curb, but the size and risk of that jump should match whether the character is an extreme athlete or a guy running late.
For adjacent camera and action techniques, see the 5 Seedance Tracking Shot Prompts for the camera-follow mechanics behind the mountain-bike and fixed-gear descents, and the 5 Seedance FPV Drone Prompts for another approach to continuous high-speed pursuit shots. The 5 Seedance Motorcycle Prompts cover the same lean-and-grip physics on a powered two-wheeler, and How to Write Seedance 2 Prompts covers the general prompt-structuring principles behind all five techniques above.
