We posted a Seedance video on TikTok and someone commented, "Where did you film this?" Nobody could tell it was AI.
The key was simple. Not "realistic video" — but "9:16 vertical music video." Telling the model the format changes everything. Here are 5 short-form prompts that actually got "this looks real" reactions.
1. K-POP Style Dance Music Video
See the full prompt on scenic.sh →
"9:16 vertical music video, low-angle medium shot, a lively stylish young woman dancing directly to camera against a clean high-saturation sky backdrop, energetic pop commercial aesthetic, fixed camera for most shots, ultra smooth rhythm and choreography, interactive performance, confident smile, playful attitude. Shot 1 (0-1s): girl in cyan top and pale yellow skirt, hands on cheeks, then points to camera, cute and lively, ending with a very fast clockwise spin transition with strong swoosh ener..."
Short-form breakdown: "9:16 vertical music video" must appear in the first line. Seedance defaults to 16:9 horizontal, and writing just "vertical" isn't enough — "vertical music video" is what shifts the entire production style to dance choreography, camera angles, and lighting designed for music video format. "Dancing directly to camera" and "interactive performance" create the fourth-wall break that makes dance videos work on TikTok. Cutting shots in 1-second intervals with specified transitions keeps energy flowing between cuts.
When writing your own music video prompts, don't skip the transition instructions. "Ending with a very fast clockwise spin transition with strong swoosh energy" controls how each shot ends, which determines whether the flow between shots holds up. A high-saturation solid-color background keeps the visual crispness TikTok's algorithm favors while preventing the dancer from getting lost in the scene.
2. Downtown Emergency — Documentary Style
See the full prompt on scenic.sh →
"15-second vertical emergency street incident, realistic cinematic documentary style, 9:16. Reference binding: @ Image1 is the main subject and must remain the same adult young woman throughout the full video. Preserve her exact face, long straight dark-brown hair with center part, fair skin, soft oval face, defined eyeliner, glossy pale pink lips, multiple silver ear studs, and yellow one-shoulder knit top with a silver oval brooch. Do not age-shift her. Do not change hairstyle, outfit, or ident..."
Short-form breakdown: "Reference binding" is the core of this prompt. One of Seedance's biggest weaknesses in longer videos is character consistency drift — faces change, outfit details vanish between shots. Explicit instructions like "must remain the same throughout the full video" and "Do not age-shift her" prevent that drift. "Documentary style" pulls in handheld camera, natural light, and ambient sound all at once — staged fiction that looks captured rather than produced is what performs best on social.
Uploading a reference image alone without character description causes appearance drift between shots. The most stable approach is image + "Preserve her exact face" + text description of outfit and hair. For social short-form, format keywords like "documentary", "raw handheld footage", "phone camera" are the most effective at erasing the AI-generated feel.
3. Surfing on an F-16 — Viral Action
See the full prompt on scenic.sh →
"aesthetic: Raw 35mm handheld, high altitude sun haze. One unbroken continuous tracking shot. No cuts. All real time. audio: Full constant jet engine roar, wind blast, no other sound.
timeline:
- 0-3s: Normal guy in baggy cargo shorts and flip flops is standing perfectly relaxed balancing on top of the wing of an F16 doing 350mph at 10,000 feet.
- 3-7s: The pilot leans out of the canopy, gives a thumbs up towards the guy on the wing. The guy leans forward slightly, smiles and returns the thum..."
Short-form breakdown: The core of virality is triggering "Is this actually real?" "Raw 35mm handheld" and "No cuts. All real time" create that believability — handheld micro-shake makes it look like someone actually filmed this, and no cuts eliminate any sense of editing. "Full constant jet engine roar, wind blast, no other sound" — ambient-only audio adds documentary texture. "Normal guy in baggy cargo shorts and flip flops" — an ordinary person in an impossible situation is the viral contrast.
For viral concepts, try the "ordinary + impossible" formula. Guy in gym clothes on a spaceship, flip flops on Everest — push the contrast between the character's normalcy and the situation's absurdity to extremes. Add "No cuts. All real time" for one-take texture, and you've got the hook that makes viewers stop scrolling and rewatch.
4. Volleyball Spike — Sports Highlight
See the full prompt on scenic.sh →
"shot": { "lens": "24mm wide-angle", "starting_position": "Camera low and to the left of the attacking player, matching the reference image exactly. Player leaping high, right arm extended toward the frozen ball above. Net runs diagonally into background.", "motion": "bullet-time 360° orbit around the frozen spike moment — slow, sweeping arc" }
Short-form breakdown: Sports highlights consistently pull high view counts on TikTok and Reels. Bullet time captures that 0.1-second spike moment with a 360-degree camera over 7-8 seconds — letting viewers experience it from angles and speeds impossible in real life. "24mm wide-angle" combined with "Camera low and to the left" makes the player look bigger and the jump higher, while "Net runs diagonally into background" ensures dynamic background shifts during the orbit.
This formula — "decisive moment + low-angle wide + bullet-time 360" — works for any sport: soccer, basketball, MMA. Each sport has its most dramatic instant. Specify it as "frozen [action] moment" and position the camera below the subject. When creating your own sports highlights, a reference image that locks the athlete's pose helps the AI nail the exact frame you want.
5. World Athletics Championships — TV Broadcast Style
See the full prompt on scenic.sh →
"Film Style: Authentic 4K Ultra-HD Global TV Sports Broadcast. Professional high-shutter speed sports photography. Visual Elements: On-screen TV graphics, 'LIVE' bug in top corner, 'WORLD ATHLETICS CHAMPIONSHIPS' scoreboard overlays. Camera Behavior: Rapid-fire professional cuts every 2 seconds. High-speed gimbal tracking and aerial helicopter shots. Color Grade: Vibrant stadium daylight, high dynamic range, crisp whites and deep track oranges. Audio: Immersive spatial sound. High-energy, breathl..."
Short-form breakdown: Starting with "Authentic 4K Ultra-HD Global TV Sports Broadcast" switches the entire frame composition, color grading, and editing rhythm to broadcast mode. On-screen UI elements like "'LIVE' bug in top corner" and "scoreboard overlays" transform a running clip into World Championships coverage. One line makes the difference. "Vibrant stadium daylight, crisp whites and deep track oranges" — pulling color grading from the location's signature colors produces the most realistic output.
This approach extends beyond sports. "Authentic Netflix documentary", "Live news broadcast", "Reality TV confessional cam" — putting a format declaration on the first line applies that format's camera style and editing rhythm wholesale. Specifying on-screen UI elements — lower thirds, logos, timecodes — eliminates the AI-generated feel and makes the output look like real broadcast footage.
Short-Form Prompt Checklist
Check these 4 things when making short-form with Seedance:
- "9:16 vertical" on the first line — Format declaration comes first. Bury it later and it may get ignored. Append a genre — "vertical music video", "vertical documentary" — to shift the entire production approach.
- Declare the broadcast/filming format — "Music video", "documentary", "TV broadcast", "handheld raw footage" — reference real-world formats. The AI imports that format's camera, lighting, and editing patterns wholesale.
- Specify camera movement physically — "Fixed camera", "handheld", "gimbal tracking", "low angle" — describe the camera's physical state. Vague terms like "dynamic" mean nothing.
- Lock character appearance — Reference image + outfit/appearance text description + "must remain the same throughout" is the combo. A reference image alone causes appearance drift between shots. The text description acts as an anchor.
Find more short-form prompts on scenic.sh.
