Hey, this is Scenic. Everyone's busy experimenting with Seedance 2.0 right now — and with Seedance, prompt quality is video quality. Same idea, completely different output depending on how you write the prompt.
We tested hundreds of prompts ourselves. Here are the 10 with the best results. All ready to copy and use.
1. Notebook Battle — 2D Hand-Drawn Animation
See the full prompt on scenic.sh →
"15 seconds, stylized 2D hand-drawn animation, overhead battlefield on aged yellow lined notebook paper, clear blue horizontal ruled lines and a red left margin line always visible, fine paper grain, pencil marks, ink strokes, minimal classroom-material aesthetic at the start. The entire video must p..."
"Aged yellow lined notebook paper" is what makes this prompt. The moment you set the background as notebook paper, every visual element becomes a drawing on that surface. "Pencil marks, ink strokes" specifies the drawing tools, preventing any 3D or digital feel from creeping in.
The "overhead battlefield" camera angle is equally important. A top-down view creates the feeling of watching doodles on a notebook. Shoot from the side and the paper texture disappears — it just becomes generic 2D animation.
2. Globe Snap-Zoom — Cinematic Montage
See the full prompt on scenic.sh →
"[CINEMATIC SETUP] Film Style: Photorealistic 8K, 50mm anamorphic. Brass desk lamp casting hard amber light from screen-left onto a walnut desk. Fast editorial montage — snap-zoom between globe surface and civilization vignettes. Camera: Locked on desk for globe shots. Snap-zoom push-in for vignett..."
"Snap-zoom" is the technique that creates this video's rhythm. Instant zoom from the globe's surface into civilization scenes, then back to the globe. This rapid back-and-forth creates that National Geographic opening feel.
"Brass desk lamp casting hard amber light from screen-left" — specifying light direction and color temperature. Seedance produces far more realistic results when you write out the light source's exact position and color rather than abstract instructions like "good lighting."
3. 90s Street Dance — Home Video
See the full prompt on scenic.sh →
"A 90s era home video, she is street dancing on a warm city street at dusk in baggy 90s clothes to an early 90s hip-hop track, a group of people are around her cheering her moves, especially when she pulls out a massive move."
This prompt is intentionally short. The single format declaration "90s era home video" determines everything — image quality, color warmth, camera shake, audio. VHS noise, warm color temperature, handheld wobble all kick in automatically.
"A group of people are around her cheering" matters too. Without an audience, it's just a dance video. With cheering bystanders, it becomes an actual street performance with real energy. Seedance handles crowd interactions surprisingly well.
4. Gothic Dark Romance — Horror/Mood
See the full prompt on scenic.sh →
"FORMAT: 15s / slow cinematic reveal / one continuous emotional escalation / no dialogue STYLE: gothic dark romance, photorealistic cinematic lighting, deep red velvet curtains, black satin fabric, premium editorial horror aesthetic, high detail, soft film grain, shallow depth of field SUBJECTS: A st..."
"One continuous emotional escalation" structures the entire prompt. Start quiet, end intense — this single line designs the emotional curve of the full 15 seconds. Without it, Seedance tends to output a flat, unchanging tone.
"Deep red velvet curtains, black satin fabric" — specifying actual materials makes Seedance simulate textures. Writing just "dark background" gives you flat black, but naming materials brings out folds, sheen, and light reflections on fabric.
5. First-Person Restaurant — POV Realism
See the full prompt on scenic.sh →
"realistic live-action, single continuous shot, full first-person POV, no cuts, no angle changes, no scene transitions, no montage, no flash cuts, no music, no subtitles. Only two characters: my unseen first-person POV and @ Image1 sitting across from me. Natural ambient sound only: light restaurant..."
A textbook example of negative instructions. "No cuts, no angle changes, no scene transitions, no montage, no flash cuts, no music, no subtitles" — listing every restriction. Seedance defaults to adding cuts, music, and subtitles, so you have to block all of them for true first-person POV.
"Natural ambient sound only: light restaurant" — specifying only ambient audio is crucial. A video without background music makes the viewer feel like they're actually there. For POV footage, sound design matters most.
6. Michelin Plating — Food Cinematic
See the full prompt on scenic.sh →
"FORMAT: 15s, cinematic pacing, continuous camera movement with focus transitions, natural transition into inner world through eye reflection, overall smooth and cohesive flow CHARACTER: @[Image 1] FOOD: Salmon rose tower, Michelin-level plating. Thinly sliced salmon layered into a rose form, with..."
"Natural transition into inner world through eye reflection" — transitioning into an inner world through the reflection in someone's eye. This kind of abstract transition actually works in Seedance 2.0. The sequence goes: eating → eye close-up → entering the reflected world.
Separating the FOOD section is essential. "Salmon rose tower, Michelin-level plating, thinly sliced salmon layered into a rose form" — this level of food description detail produces results virtually indistinguishable from actual Michelin-level dishes.
7. Figure Skater — Ice Ballet
See the full prompt on scenic.sh →
"SUBJECTS: A top-level female figure skater in an ice ballet style, blue eyes, gaze focused and steady, with a slight flush from the cold; Hair styled in a clean, neat low bun, with a few loose strands subtly moving with motion; Wearing a figure skating costume with a white to light pink gradient and..."
"A slight flush from the cold" and "a few loose strands subtly moving with motion" — these micro-level physical details are what make AI video feel real. Cheeks slightly reddened from cold, a few stray hairs moving with each movement. The AI won't add these unless you explicitly write them.
Color gradients in the costume matter too. "White to light pink gradient" instead of just "white costume" produces something that shifts with on-ice lighting, looking far more premium.
8. Roller Coaster Ride — In-Camera Inset
See the full prompt on scenic.sh →
"FORMAT: 15s / 1 shot / live rear-follow in-camera inset SUBJECTS: One blonde female rider in a low-cut flowing white dress under a locked leather coaster harness, seen from behind in the main live frame while a small bordered rectangular top-right in-camera inset keeps her visible from the waist up..."
"In-camera inset" puts a screen within a screen. The main frame follows from behind while a small top-right window shows a front-facing close-up. The fact that Seedance handles this split-screen from a prompt alone is impressive.
"Flowing white dress under a locked leather coaster harness" — fabric billowing in the wind while being anchored by a safety harness. Writing these physical contradictions together ("flowing yet restrained") makes Seedance simulate both, increasing realism.
9. Creative Director — Era Jumps
See the full prompt on scenic.sh →
"CHARACTER: attached image — confident creative director, late 30s, slim build, short dark hair, natural wide smile; walks with calm executive energy; outfit morphs each jump to reflect the creative era CAMERA: frontal tracking, eye-level, full-body, dolly backward with character, never cuts away fro..."
"Outfit morphs each jump to reflect the creative era" — one character walks forward while the background and outfit shift through eras. 60s → 80s → modern, background and wardrobe transforming together while the camera never breaks from a frontal tracking shot. Classic fashion brand ad technique.
"Dolly backward with character, never cuts away" — the camera moves backward, keeping the character centered in frame. With this instruction, even as eras change, the viewer's focus stays on the character, maintaining the "one person's journey" narrative.
10. 90s Kitchen Fight — One-Take Action
See the full prompt on scenic.sh →
"[CINEMATIC SETUP] Film Style: 1994 high-budget action-crime aesthetic, 35mm film grain, rich cinematic texture. Lens: 24mm wide-angle anamorphic for a deep field of view during the long take; sharp focus. Color Grade: Contrast between cold fluorescent kitchen lighting (teals/whites) and warm, golden..."
"1994 high-budget action-crime aesthetic" — combining a year with a genre makes Seedance import that era's film texture, color grading, and camera patterns wholesale. Just writing "cinematic" lets the AI interpret freely; "1994 action-crime" narrows it to reference points like The Perfect Storm or The Rock.
"24mm wide-angle anamorphic" is an actual lens specification. Seedance responds to this technical level by simulating lens distortion, horizontal flares, and anamorphic bokeh. Write just "wide lens" and all that detail vanishes.
Prompting Tips
Patterns that repeat across all 10:
- Declare the format first — "90s home video", "1994 action-crime", "2D hand-drawn animation" — the format declaration determines everything about the video.
- Negative instructions are powerful — "No cuts, no music, no subtitles" — explicitly stating what not to do breaks the AI out of its default habits.
- Micro physical details create realism — "Slight flush from the cold", "loose strands moving with motion" — these details close the gap between AI and real footage.
- Specify the light source — Instead of "good lighting," write "brass desk lamp casting amber light from screen-left."
Find more prompts at scenic.sh.
