Seedance 2.0 Shot Design Workflow

The shot design workflow follows five sequential steps: (1) Requirement Analysis — confirm duration, aspect ratio, generation mode, style direction, and any reference material; (2) Visual Diagnosis — select a director style preset, plan the storyboard structure, and decide whether a single segment or multi-segment approach is needed; (3) Six-Element Precision Assembly — build the complete prompt using the six structural elements (subject, action, scene, style/lighting, camera/focal length, sound); (4) Validation — review the assembled prompt against a quality checklist, checking for filler words, missing elements, and potential content filter triggers; (5) Delivery — output the final prompt (or multi-segment prompt set) in the correct format for the target platform. Each step has specific decision points and can loop back to earlier steps when the validation check reveals issues.

Before writing any prompt, lock down the production parameters. Duration determines whether you need a single segment (≤15 seconds) or multi-segment storyboard (>15 seconds). Aspect ratio (16:9, 9:16, 1:1, or 4:3) affects framing decisions and camera movement options. Generation mode — text-to-video, image-to-video, or video-to-video — changes which elements the prompt must describe versus which ones the reference files handle. Gather any reference material early: character sheets, style frames, camera motion clips, or audio tracks. Write down the creative brief in one sentence before proceeding: 'A 30-second Xianxia fantasy clip in 16:9 with sword combat and ink-wash transitions.' This anchors every subsequent decision.

Choose a director style preset that matches the creative brief. The preset acts as a visual anchor that keeps all downstream decisions consistent. For epic sci-fi, you might select Villeneuve (vast desolation, geometric framing, desaturated palette). For an emotional character piece, Wong Kar-wai (step-printed slow motion, saturated color, handheld intimacy). For Eastern fantasy, the Xianxia preset (aerial temples, ink-wash transitions, ethereal audio). You can combine presets — 'Villeneuve composition with Wong Kar-wai color' — for hybrid identities. Next, sketch the storyboard: how many shots, what happens in each, where the emotional peak lands, and how segments hand off to each other. For clips under 13 seconds, a single shot description suffices. For 13–15 seconds, use timestamped storyboard format. For anything longer, plan multi-segment splits.

Every professional prompt should cover six elements in order: (1) Subject and Appearance Details — define the character or object, including physical build, costume, key visual identifiers, and emotional state; (2) Action and Physics Continuity — describe what happens with physically plausible motion and continuity constraints; (3) Scene Environment — set the location, time of day, weather, spatial depth, and atmospheric conditions; (4) Visual Style and Physical Lighting — apply the director style preset and build the three-layer lighting structure (source, behavior, tone); (5) Physical Focal Length and Camera Movement — specify the lens (24mm wide, 85mm portrait, 200mm telephoto) and the camera motion (dolly tracking shot, crane shot, slow orbit, handheld); (6) Native Sound Effects — describe material-specific sounds and spatial acoustic modifiers. This formula extends the SCELA framework by promoting sound from an optional afterthought to a first-class structural element.

Lighting is one of the most impactful elements in video generation, but it is frequently described too vaguely. The three-layer structure forces precision. Source Layer: name the actual light sources ('two warm practicals flanking the doorway plus cold moonlight from the window'). Behavior Layer: describe how light interacts with materials and atmosphere ('volumetric fog catching the moonlight beams, soft specular highlights on wet floor tiles, shadow edge diffusion on fabric'). Color Tone Layer: define the palette ('cool blue dominant with warm amber accents in the foreground, teal shadow tones'). A concrete example for a Xianxia scene: 'Source — golden sunset backlighting through temple columns plus floating ethereal orbs; Behavior — dust motes diffusing the backlight, silk robe catching soft specular highlights; Tone — warm amber base transitioning to cool jade at the edges.' This level of specificity produces dramatically more consistent lighting than 'cinematic lighting' or 'dramatic atmosphere.'

For clips between 13 and 15 seconds, timestamp storyboarding gives the model explicit timing cues for each beat. The format opens with a style overview, then assigns time ranges to visual and camera instructions: '[Style overview]. 0–3s: [establishing shot, wide angle, camera pulls back to reveal environment]. 3–8s: [subject enters frame, medium shot, slow dolly forward]. 8–12s: [close-up on key action, shallow depth of field, subtle camera shake]. 12–15s: [resolution beat, wide shot, camera rises slowly]. Lighting: [three-layer description]. SFX: [material-specific sound description]. Negative: any text, subtitles, logos or watermarks.' Each time block should contain one clear action and one camera setup. Avoid packing multiple actions into a single 3-second window — the model handles one deliberate movement per beat more reliably than rapid multi-action sequences.

When the target duration exceeds 15 seconds, split the video into self-contained segments using the formula: number of segments = ⌈total_duration / 15⌉, with the final segment being at least 8 seconds long. A 40-second video becomes three segments (15s + 15s + 10s), not four with a 10-second tail. Each segment is a complete, independent prompt with timestamps starting from 0, but all segments share: a unified style preamble (director preset and visual identity), consistent three-layer lighting (which may evolve narratively but stays within the established palette), stable handoff frames at segment boundaries (the last 1–2 seconds should land on a composable resting state — freeze, slow push, or fade), and a shared forbidden-items declaration. Distribute the narrative arc across segments: Segment 1 is the opening (establish world, introduce subject), middle segments develop the story (action, conflict, escalation), and the final segment delivers the climax or resolution.

The most common prompt weakness is relying on abstract filler words that sound impressive but give the model no visual target. Words to eliminate: 'masterpiece,' '4K,' '8K,' 'ultra HD,' 'ultra-clear,' 'best quality,' 'amazing,' 'stunning.' These are quality claims, not quality descriptions. Replace them with physical material anchors: instead of 'ultra-clear,' write 'Kodak 5219 film stock warmth with fine grain visible on skin tones'; instead of 'high quality textures,' write 'brushed aluminum with micro-scratches catching rim light, wet concrete reflecting neon, aged leather with cracked patina.' Include organic imperfections that signal realism: lens dust, subtle focus breathing, micro camera shake, film gate weave. A prompt filled with specific material descriptions consistently outperforms one padded with superlatives.

When a creative brief references recognizable intellectual property, use the three-tier fallback to avoid content filter rejections. Level 1 — Name Replacement: swap the IP name for an original descriptive nickname that captures the archetype ('Iron Man' becomes 'Alloy Sentinel,' 'Spider-Man' becomes 'Web-Strand Acrobat'). Level 2 — Feature Modification: replace iconic visual traits — signature color schemes, costume silhouettes, weapon designs, logo elements — with original alternatives that preserve the narrative function. Level 3 — Category Abstraction: remove all visual and nominal connections to the original IP, keeping only the role ('a powered-armor hero' instead of any Iron Man derivative). Always add explicit forbidden items for words that might trigger the filter. Start at Level 1 for each character; if the platform still flags the prompt, escalate to Level 2, then Level 3. Document which level worked for each IP concept so the team can reuse the mapping.

Seedance 2.0's native audio generation responds to physics-based sound descriptions far better than generic music terms. Structure sound descriptions by category: ambient sounds set the environment (wind whistling through temple corridors, rain pattering on a metal roof, distant thunder rolling across mountains), action sounds punctuate events (blade cutting through air, boots crunching on gravel, glass shattering in slow motion), vocal elements add human presence (whispered dialogue with breath condensation, crowd murmur fading to silence), and material-based onomatopoeia adds texture (silk rustling against stone, metal scraping on metal, wood creaking under weight, water dripping onto slate). Add spatial acoustic modifiers to describe the sound environment: cathedral reverb for large enclosed spaces, tight room dampening for intimate scenes, outdoor open-air for nature sequences, underground echo for cave or tunnel settings. The combination of material-specific sounds and spatial modifiers produces immersive audio that reinforces the visual scene.