1. First, what is Seedance 2.0—and why use it?
Seedance 2.0 is ByteDance’s new-generation AI video generation model, tuned for Chinese-speaking creators, and it shortens the path from “idea in words → frames on screen.” Roughly, the main strengths are:
- Dual modes: text-to-video (a paragraph becomes a clip) and image-to-video (a still becomes a moving shot).
- Motion quality: Actions tend to stay coherent; “broken faces / hands” happen less often than with early models, and motion reads better.
- Detail & physics: Water, cloth, light shifts, and similar cues are easier to render with nuance.
- Chinese prompts: Native understanding of Chinese descriptions—less “translate first, then reroll.”
- Broad styles: Realistic, anime, 3D, ink-wash, cyberpunk, and more (subject to what the current build supports).
In one line: among Chinese AI video tools, it often saves effort on motion quality and prompt understanding, so beginners can get a usable clip sooner. Quotas, feature entry points, and compliance follow the official product and docs.
2. How do you write prompts that actually work?
Get the prompt right and you’ve already won half the battle. Use one golden formula as a backbone, then fill six dimensions—skip one and the frame often drifts.
2.1 Golden formula
Subject + specific action + scene detail + camera setup + light & mood + style & texture
Each layer should be concrete and visualizable; avoid empty adjectives.
2.2 Six dimensions (good vs bad)
① Subject
- ❌ “A girl”
- ✅ “Energetic teen with a high ponytail, white dress, soft smile”
② Action (verb + scale / rhythm)
- ❌ “The girl runs”
- ✅ “Runs holding her hem, skirt lifts slightly with each step, light brisk pace”
③ Environment (time + place + visible detail)
- ❌ “Outdoors”
- ✅ “Early morning meadow, pale purple wildflowers, distant green hills, thin mist mid-air”
④ Camera (shot size + movement)
- Shot size: wide, medium, close-up, etc.
- Movement: lock-off, slow push-in, follow, orbit.
- ✅ “Medium shot, slow push-in along the run direction”
⑤ Light & mood (light type + contrast + color emotion)
- ❌ “Nice light”
- ✅ “Soft morning sun on the grass, warm yellow key, dappled patches, fresh healing vibe”
⑥ Style & texture (look + clarity / grain you want)
- ✅ “Anime look, clean lines, moderate saturation, clean frame, low noise”
2.3 Average vs strong: same brief
Average: Girl dances in a park with pretty light
Strong: Dance student, low ponytail, black practice wear, ballet on a leaf-strewn park path—single-foot spin, skirt arcs into a circle; wide shot, slow orbiting move; sunset sidelight, golden rays, dappled leaf shadows; photoreal, high resolution, crisp detail.
2.4 Camera cheat sheet
| Keyword (CN-friendly in product) | Effect | Good for |
|---|---|---|
| Close-up | Face or object detail | Expressions, food, product |
| Medium shot | Waist-up, person + some context | Dialogue, daily life |
| Wide shot | Full space | Landscapes, architecture, scale |
| Aerial / top-down | Bird’s-eye | Cities, nature |
| Follow | Moves with subject | Run, walk, chase |
| Orbit | Circle subject | People / product showcase |
| Slow push-in | Far to near | Tension, reveal |
| Low angle | Camera looks up | Power, hero shot |
| POV / first person | Subjective view | Immersion, exploration |
| Time-lapse | Compress time | Sunrise/sunset, traffic |
2.5 Rhythm control words
| Phrase cluster | Rhythm | Examples |
|---|---|---|
| Slowly, gently, unhurried | Slow, elegant | Petals falling, glance back, veil |
| Fast, suddenly, rapid | Fast, punchy | Sprint, lightning, blast |
| Gradually, ease into | Smooth transition | Sky shift, pose change |
| Suddenly, abruptly | Shock, tension | Door burst, hard cut to CU |
| Rhythmic, groovy | Repeated order | Dance, waves, robot arm |
3. The “killer” mode: image-to-video
Image-to-video is often the most expressive—and the easiest place to pull ahead. Pick the right image + describe motion correctly so the still “comes alive” naturally.
3.1 Five rules for choosing images (your floor)
- Resolution first: A blurry still rarely becomes a sharp video; aim for ~1024px+ width; more pixels usually mean stabler detail in motion.
- Clean composition: Clear subject, less clutter—fewer glitches and weird merges.
- Starting pose: The pose in the image ≈ frame one; pick a key pose so the follow-through flows.
- Light depth: Side or back light with contrast carries mood into motion more easily.
- Minimal text: Letters warp and smear in video—prefer text-free plates.
Compliance: Do not upload images that violate portrait rights, copyright, or platform rules; follow Jimeng (Dreamina) terms and moderation.
3.2 Where assets come from (ideas)
| Source | Traits | Typical use |
|---|---|---|
| Midjourney | Strong aesthetics & mood | Cinematic characters / epic scenes → I2V |
| Flux | High prompt adherence | When you need exact detail / pose lock |
| Jimeng (Dreamina) | Same ecosystem, smooth handoff | Often pairs cleanly with Seedance motion |
| Real photos | Authentic detail | Portraits, landscapes, products |
| Original art | Distinct look | Anime shorts, picture-book style, experiments |
3.3 Image-to-video prompt rules
Don’t restate what’s already frozen in the image; focus on change, action, motion, camera.
- ❌ “A girl in white under cherry blossoms” (already in frame—confuses the model)
- ✅ “Breeze lifts hair and hem, petals drift down, she lifts her chin toward the sky, fingertips slightly raised; slow push-in, overall motion soft and natural”
4. Ten practical playstyles (method + sample prompt)
Play 1: Cinematic short
Method: Write 4–8 shots as separate generations; end each prompt with a shared style anchor (e.g. filmic, warm grade, shallow DOF, 35mm grain); edit with dissolves + score.
Beat examples
- Open: Misty alley, aerial descent, morning light through old locust, warm film look.
- Character: Woman in pale qipao eases a wooden door; medium, rim light.
- Walk: Oil-paper umbrella, stone path, side tracking, hem sway.
- End: She pauses on a stone bridge, camera pulls back to reveal the old town, long aftertaste.
Play 2: Premium product ad
Method: Start from a sharp product still; don’t repeat static product traits in text—push spin / lift motion + light + camera.
Sample: Frosted silver bottle rotates slowly on axis, black gradient backdrop, cool key from top-side, fine metal highlights, thin mist shifts at the base; orbit move, slow pace, low-sat dark grade, minimal luxury.
Play 3: Old photos that “move”
Method: Pick storytelling vintage shots; restore quality if needed; prompt micro-motion + keep grain / vintage color, avoid extreme warp.
Sample: In a B&W group portrait, elders blink lightly, kids’ lips barely lift, hair trembles, old locust leaves sway slowly; keep grain and mono, subtle film weave. For single color portraits, warm vintage cast, gentle sway, slow blink.
Play 4: Nature time-lapse
Method: High-res landscape plate; spell out time passage, cloud/wave/light change + camera scale.
Sample: Rocky coast time-lapse, sun rises, pink light on the sea, waves in rhythm; low aerial wide, documentary contrast. Forest sea of clouds / prairie sunset—swap subject and light words with the same logic.
Play 5: Anime / ACG short
Method: Detailed character or scene art; stress 2D motion + color + light mood.
Sample: Blue twin-tails magical girl, wheat under stars, crystal staff glow, skirt and wheat in wind, star dust falls; clean anime lines, purple-blue gradient, soft bloom, healing mood.
Play 6: Creative loop clip
Method: Design loopable motion for short-form.
Sample: Lime slice slowly rises spinning in sparkling water, bubbles rise and pop from the bottom; top-down close-up, cool natural light, clean BG, smooth loop, crisp ASMR feel.
Play 7: Chinese ink style
Method: Ink base art; describe ink bleed, robe motion, mist over mountains.
Sample: White-robed swordsman in mountains, robe and tassel sway, distant peaks emerge in fog, ink spreads at edges; negative space, black-gray-white with a touch of ochre, rice-paper texture, airy mood.
Play 8: Lifestyle realism
Method: Simple story + reference still; lock interaction + emotion.
Sample (afternoon tea): Sunny balcony, two friends at a wood table—one forks strawberry cake, one lifts a teacup to clink, they smile at each other; warm natural light, wood interior, medium follow, photoreal detail.
Play 9: Arch / interior viz
Method: Render or layout as base; describe travel path + curtain / light micro-motion.
Sample: Slow push from entry through marble hall, curved floor-to-ceiling glass and skyline, sheer curtains drift, soft sun on pale gray sofa; modern luxe, cool gray palette, magazine detail.
Play 10: Meme-cute short
Method: Contrast + exaggerated cute motion, fast and punchy.
Sample: Round corgi in tiny suit on a bench, paws on mini coffee, licks lips, proud eyes; pigeons strut behind; medium, bright daylight, slightly exaggerated comedy.
5. Three advanced moves (double the polish)
Move 1: Style lock
End every shot with the same style anchor so grade, grain, and aspect mood stay unified. Build templates per project, e.g.:
- Cinematic: Filmic, shallow DOF, light grain, warm orange-cream grade, wide feel, subtle vignette.
- Anime: Cel look, clean line, macaron saturation, soft bloom, light diffusion, low noise.
- Cyberpunk: Neon highlights, teal-magenta clash, wet street bounce, high contrast, cold metal.
- Vintage tape: Faded warm yellow, mild cast and distortion, light VHS noise (use sparingly).
Move 2: Negative prompts (if supported)
State what you don’t want to cut broken faces, extra fingers, random text, heavy shake, clipping, color mud, etc. Keep a checklist per project, e.g.: blur, twisted limbs, wrong finger count, on-screen text/watermark, violent shake, edge clipping, dirty color, heavy noise, blown highlights or crushed blacks.
Move 3: Generate many, pick one
Same prompt still has variance. For hero shots, run 3–5 takes and score motion smoothness, detail, and emotion fit; B-roll can use fewer passes to save time.
6. Common failures & fixes
| Symptom | Likely cause | Fix |
|---|---|---|
| Face warp | CU too tight or expression overload | More medium shots; simple stable expressions |
| Hand issues | Industry-wide weak spot | Hands secondary or hidden by props; crop in post |
| Disjointed motion | Too many actions in one prompt | One core action per clip; split and edit |
| Ignores prompt | Prompt too long / buried lead | ~80–150 Chinese characters ballpark; key info first, style last |
| Style mismatch | No shared anchor | Same closing style line every time |
| Shot feels cut inside one gen | Hard scene/angle jump in one prompt | One scene per generation; use editing for cuts |
Watermarks & post: If exports include platform marks, follow the terms first; only use owned or licensed assets and compliant post when rules allow—avoid infringement.
7. Workflows & habits (straight talk)
Fast pass (~minutes)
Prompt → text-to-video → same settings, several rolls → pick best.
Polished single (~half hour)
T2I or Jimeng for a sharp plate → image-to-video → grade, music, captions in edit.
Full short (1–2 hours)
Shot list → shot-by-shot gen (locked style) → edit → transitions, music, SFX, subs → overall grade → export.
Subjective vs other tools: Seedance 2.0 often wins on Chinese semantics + motion; outcome still depends on subject, reference quality, and iterations—A/B with the same prompt.
Seven long-term habits
- Drop “first try must be perfect”—compare multiple for key shots.
- Build a prompt template library by style/scene; swap only subject/action.
- Beginners: subtract—one clear action + one clear scene; add detail after it’s stable.
- Core of AI video is motion: who moves, how, at what rhythm.
- For series, lock the style anchor for a recognizable look.
- Watch blog updates; test new features early.
- Mix genres and niche ideas within compliance to expand your range.