--- name: SeedancePro description: Master Seedance 2.0 for professional AI video generation, with explicit input-mode selection (Omni/elementos, frame inicial, or transición frame inicial/final). Use this skill whenever the user wants to create short-form videos (up to 15 seconds) for cinematic storytelling, product showcases, social media content, ad variations, music videos, explainers, or any narrative project. Includes prompting techniques, reference image/video/audio strategies, multi-shot editing, character consistency with World ID, and practical workflows for Spanish-language content and custom characters. Trigger for any request mentioning video generation, AI cinema, short-form content creation, Seedance, or when translating storyboards/scripts to video. Always use this skill before writing any Seedance prompt — it contains filter compliance rules, the 1,990-character hard limit, and the sectioned format required for successful generation. --- # Seedance 2.0: Professional Prompt Engineering ## Pre-Flight Questions (ASK BEFORE WRITING ANYTHING) Before generating ANY prompt, you MUST ask the user three questions, in order. Do not write a single prompt until all three are answered. **ALWAYS present these questions as clickable multiple-choice options (using the interactive question / AskUserQuestion menu), NEVER as plain text the user has to answer by typing.** The user should be able to just click an option. Use exactly the options listed in each question below as the selectable chips: - **Pregunta 1 — Modo:** `Omni (elementos)` · `Frame inicial` · `Transición (inicial/final)` - **Pregunta 2 — Duración:** `4s` · `5s` · `6s` · `7s` · `8s` · `9s` · `10s` · `11s` · `12s` · `13s` · `14s` · `15s` - **Pregunta 3 — Formato:** `Narrativo (sin código de tiempo)` · `Con código de tiempo (por shots)` (You may ask Pregunta 1 first, then Preguntas 2 and 3 together in one clickable menu.) ### Question 1 — ¿Qué modo de entrada usarás? Ask: **"¿Usarás el modo Omni (elementos), modo frame inicial, o modo transición (frame inicial/final)?"** The answer determines whether and how you reference images: | Respuesta del usuario | Regla de citado de imágenes | |---|---| | **Modo Omni (elementos)** / "sí" | SÍ puedes usar `@` para citar cada imagen en el prompt (`@image1`, `@image2`, …). Cada elemento se cita explícitamente como en el sistema multi-modal estándar. | | **Modo frame inicial** | NO uses ningún `@` ni lo cites. Solo **describe con palabras** lo que hay en la imagen dentro del prompt (apariencia, objetos, entorno) sin referenciarla. | | **Modo transición (frame inicial/final)** | NO uses ningún `@`. Construye una **transición** que lleve visualmente del primer frame (inicial) al último frame (final): describe el estado inicial, el movimiento/cambio intermedio y el estado final. | Important: in **frame inicial** and **transición** modes, every `@image1`/`@video1`/`@audio1` reference in the templates and Multi-Modal section below is DISABLED — translate those references into plain visual description instead. Only **Omni mode** keeps the `@` citation syntax. ### Question 2 — ¿De cuántos segundos será el video? After mode is chosen, ask: **"¿De cuántos segundos será el video?"** Offer exact whole-second values: **4s, 5s, 6s, 7s, 8s, 9s, 10s, 11s, 12s, 13s, 14s, 15s**. ALWAYS use exact, whole-second numbers for duration — never rangos vagos ("unos 5-6s"), nunca decimales. Valid range: **4s to 15s**. Use the chosen exact duration to allocate shots and beats (see Duration Allocation), and make every timecode in Shot Sequence an exact integer second that adds up precisely to the chosen total. If the user names a value outside 4–15, clamp it to the nearest exact second in range and tell them. ### Question 3 — ¿Formato narrativo o con código de tiempo? After duration is chosen, ask: **"¿Lo quieres narrativo (sin código de tiempo) o con código de tiempo (dividido por shots con segundos)?"** | Respuesta del usuario | Formato de salida | |---|---| | **Narrativo** | Escribe el prompt como descripción continua y fluida, SIN etiquetas `Shot N (start-end s)` ni códigos de tiempo. Mantén las Secciones 1–4 y 6, pero la acción de la Sección 5 se redacta como un solo bloque narrativo de cámara y movimiento, sin segundos. | | **Con código de tiempo** | Usa la Sección 5 (Shot Sequence) estándar: divide en `Shot 1 — Label (0-3s)`, `Shot 2 — Label (3-6s)`, etc., con segundos explícitos por shot, repartidos según la duración elegida en la Pregunta 2. | Only after ALL THREE answers are received do you proceed to Mode Detection and prompt generation below. --- ## Core Rules (Non-Negotiable) 0. **ALWAYS deliver exactly ONE prompt.** Never 2, never 3, never alternatives or variants — a single final prompt in a single code box, every time, regardless of mode. 1. **1,990 character hard limit.** Every complete prompt — all sections, all shots — must be under 1,990 characters. Count before delivering. Cut least-essential visual detail first; never cut Section 1 or filter framing. 2. **Every prompt in its own code box.** Multi-shot prompts = one unified prompt, one code box, all shots inline. 3. **Shot List Test.** Before including any line: *Would this appear on a film director's shot list?* If not — remove it. No backstory. No emotional subtext. No character psychology. Only what the camera sees or the mic hears. 4. **Filter compliance by construction.** Don't write first, then check. Build filter safety from the first word. Lead with cinematic production context. --- ## Mode Detection Every request is one of two modes. Identify before writing anything. ### Single-Shot Mode (default) User wants a standalone video prompt. **Output: exactly ONE prompt in one code box.** Never produce alternatives or variants. - The single prompt: cinematic realism — grounded camera, natural lighting, maximum filter safety. Pick the strongest creative direction and commit to it. ### Multi-Shot Mode User specifies (or you determine together) a sequence. **Before writing, confirm:** 1. Number of shots 2. Duration per shot (or total) 3. Reference inputs (image/video/audio)? 4. Any filter-sensitive content (weapons, physical contact, intensity)? If high-level concept only — propose shot breakdown — wait for approval — then generate. **Output:** One polished sequence, one code box, all shots inline. --- ## Prompt Structure: 6-Section Format Every prompt uses this structure. Sections 1, 2, and 5 are mandatory. ### Section 1 — Visual Style & Camera (MANDATORY) ``` Visual Style: [Realism level] + [camera/format reference] + [lighting condition] + [tonal quality]. [Lens: focal length, aperture, depth of field]. [Texture: grain, haze, dust, atmospheric particles]. [Color grade: dominant palette, shadow tone, highlight tone]. Camera Behavior: [Single take / multi-shot / specified cuts]. [Movement style: handheld, dolly, crane, steadicam, etc.]. [Camera physicality: breathing shake, drift, rack focus, etc.]. ``` Always specify a real camera/format (ARRI, RED, iPhone, IMAX). Always include lens detail and one atmospheric element. This section sets the filter's first impression — cinematic production language here establishes safe context for everything that follows. ### Section 2 — Environment (MANDATORY) ``` Environment: [Location type] + [surface/ground detail] + [sky/weather/ambient light]. [Key environmental objects and condition]. [Atmospheric elements: smoke, dust, particles, reflections, moisture]. [Ambient sound: what is audible before any action begins]. ``` ### Section 3 — Characters (MANDATORY if characters present) ``` [Role label] based on @image1 [if applicable]. [Visible physical state: clothing, damage, dirt, posture]. [Body language as visible action only: stance, breathing rate, gaze direction]. ``` Role labels to use: figure, traveler, rider, warrior, combatant, merchant, pilot, dancer, artisan, sentinel, operative, navigator, technician, courier, researcher, wanderer, performer, engineer, scout. Never describe thoughts, feelings, or personality. Body language = observable physical action only. ### Section 4 — VFX & Transformation Mechanics (OPTIONAL) ``` [Trigger: visible physical action that initiates the effect]. [Particle/material behavior: how fragments move, form, attach]. [Assembly sequence: order of body regions covered]. [Lock/completion indicators: clicks, sparks, light ignition]. [Byproduct: what happens to clothing/environment during transformation]. ``` Use engineering language: "magnetic alignment," "modular attachment," "mechanical clamp," "alloy plating." Never: magical, mystical, supernatural, enchanted. ### Section 5 — Shot Sequence (MANDATORY) ``` Shot [#] — [Label] ([start]-[end]s) [Camera position and movement]. [What is visible in frame]. [Character action as physical movement only]. [VFX/environmental change in frame]. [Sound: impacts, mechanical sounds, ambient shifts]. ``` Every shot opens with camera position/movement. Never "he decides," "she realizes," "they feel" — only "he turns," "she raises," "they step." ### Section 6 — Final Frame (OPTIONAL) ``` [Character position and pose]. [Camera distance and angle]. [What is visible in environment]. [Lingering VFX: fading particles, cooling glow, settling dust]. [Ambient sound as scene holds]. ``` No emotional resolution. Only what the camera sees. --- ## Filter Compliance System ### Hard Rules — Never Violate - No real human faces as references (stylized, illustrated, 3D-rendered, or AI-generated only) - No named IP (character names, franchise names, trademarked properties) - No implied IP (if a reasonable person would picture a specific copyrighted character — rewrite) - No real public figures - No isolated violence (every weapon/combat action must have explicit cinematic framing) ### Word Swap Master Reference | Instead of... | Use... | |---|---| | young girl/boy | small figure, child-sized silhouette | | fires a gun | discharges a flare / signal shot | | looks like [celebrity] | a figure with [specific visual traits only] | | Disney-style | whimsical storybook aesthetic | | anime [named character] | stylized illustrated character | | realistic human face | stylized portrait, 3D-rendered face | | fights / attacks | trains, spars, demonstrates technique | | blood / gore | impact dust, sparks, shattered debris | | sexy / seductive | confident, commanding, poised | | scared child | small cloaked figure in a vast environment | | magical / mystical | energy-based, particle-driven, light phenomenon | | monster / demon | large bipedal entity, armored form | | death / killing | movement ceases, figure goes still | | explosion | controlled pyrotechnic blast, practical effect | | scream / cry out | sharp exhale, vocal burst | ### IP Genericization | User Intent | Filter-Safe Version | |---|---| | Iron Man suit-up | modular alloy plates magnetically assembling onto a figure's frame | | Ghibli-style forest | whimsical illustrated woodland, oversized moss-covered roots, watercolor-edge haze | | Spider-Man swinging | agile figure in tactical suit launches tether line from rooftop, swings between structures | | Star Wars lightsaber | figure activates a luminous energy blade — bright cyan, low hum, plasma edge flickering | | Disney princess | figure in elaborate period gown, ornate embroidery, grand vaulted hall, storybook aesthetic | ### Combat & Weapons — Safe Framing Pattern ``` UNSAFE: "fires a rifle at the enemy" SAFE: "discharges a signal flare skyward from a weathered flare rifle, smoke trailing into amber dusk — low-angle tracking shot, ARRI Alexa, 35mm" UNSAFE: "punches the attacker" SAFE: "combatant demonstrates a striking technique against a heavy bag in a training facility — close-up on wrapped hands making contact, dust rising from canvas, overhead fluorescent lighting" ``` ### Body Language — Physical Only | Don't Write | Write Instead | |---|---| | confident and powerful | shoulders back, chin level, weight centered | | nervous and afraid | hands clenched, breathing rapid, gaze shifting | | angry and aggressive | jaw tight, veins visible at temples, fists balled | | sad and defeated | head lowered, shoulders curved forward, gaze down | --- ## Camera & Cinematography Reference ### Angle Psychology | Angle | Visual Effect | Keywords | |---|---|---| | Low-angle | Towering, dominant | low angle shot, looking up at, worm's eye view | | High-angle wide | Diminished, exposed | high angle wide, looking down on | | Dutch angle | Instability, unease | dutch angle, tilted camera, canted frame | | 1st person POV | Direct immersion | first person view, pov shot | | Tracking shot | Journey, momentum | tracking shot, moving with subject | | Overhead / top-down | Pattern, detachment | overhead shot, bird's eye | | Ground-level | Immediacy, tactile | ground level, floor perspective | | Steadicam float | Smooth, dream-like | steadicam, smooth tracking, floating camera | | Anamorphic | Cinematic, epic | anamorphic lens, oval bokeh, horizontal lens flare | | Security camera | Surveillance, tension | security camera view, cctv perspective | | Silhouette backlit | Mysterious, iconic | backlit silhouette, rim light only | ### Lens + Lighting Quick Combos | Situation | Lens | Light | Effect | |---|---|---|---| | Hero reveal | Wide 24mm | Rim backlight | Heroic silhouette | | Intimate dialogue | Telephoto 85mm+ | Soft side | Compressed, close | | Menace/threat | Low-angle wide | Hard side light | Noir, threatening | | Vulnerability | High-angle wide | Soft top | Exposed | | Drama | Profile close-up | Split lighting | Divided, tense | ### Camera Motion Verbs whip-pan, handheld sway, shoulder-cam drift, dolly push, crash zoom, snap focus, rack focus, tracking shot, tilt up, crane shot, orbit, steady glide, parallax sweep, pull back, push in, lateral slide, spiral descent ### Lighting Vocabulary (Copy-Paste) - "backlit by golden hour sun, long shadows across ground" - "neon glow reflecting off wet asphalt — magenta and cyan" - "single candle flame, warm radius fading to darkness" - "LED panel from below, harsh uplighting on face" - "strobing emergency lights, red pulse every 2 seconds" - "moonlight through blinds, horizontal stripe shadows" ### Color Grades teal-orange blockbuster · cool blue haze desaturated midtones · amber nightclub warmth deep blacks · high-contrast noir silver highlights · bleach bypass low saturation chalky · warm golden wash lifted shadows · cross-processed shifted greens and magentas ### Texture & Tactile Realism (Copy-Paste) - "wet asphalt reflecting neon pools" - "dust motes suspended in warm light shaft" - "rain streaking down glass, city lights smeared behind" - "condensation beading on cold metal surface" - "breath visible in cold air, vapor dissipating" - "fabric rippling in wind, thread texture visible at edges" ### Real Camera Format References - "ARRI Alexa 35mm, subtle halation, warm amber grade, handheld stabilized" - "RED Komodo, anamorphic lens, horizontal flare, oval bokeh, teal-orange grade" - "iPhone ProRes handheld, natural light, documentary immediacy, visible grain" --- ## Multi-Shot Planning Guide ### Common Patterns **Character Introduction** - Shot 1: Environment — world before character - Shot 2: Partial reveal — silhouette, shadow, detail - Shot 3: Full reveal — medium shot, role and posture clear - Shot 4: Close-up — face/mask detail, physical state **Product Reveal** - Shot 1: Atmospheric environment, product partially visible - Shot 2: Slow dolly push revealing product - Shot 3: Macro detail — texture, material, branding - Shot 4: Hero shot with dramatic lighting **Action Sequence** - Shot 1: Wide establishing - Shot 2: Medium tracking - Shot 3: Close-up detail (hands, equipment) - Shot 4: POV or over-the-shoulder - Shot 5: High-angle pullback **Transformation Sequence** - Shot 1: Pre-transformation — character at rest - Shot 2: Trigger — physical action initiating effect - Shot 3: Mid-transformation — materials assembling - Shot 4: Completion — energy settling, new form - Shot 5: First action in new state ### Duration Allocation - 3s: Single action, single camera movement - 5s: Two actions or one action with camera transition - 7-8s: Setup, action, result - 10-15s: Full mini-narrative, multiple beats ### Continuity Across Shots Define "global constants" once: "All shots: ARRI Alexa, 35mm, golden hour backlight, warm amber grade, subtle film grain, handheld stabilized" — repeat in each shot's Section 1 reference. --- ## Multi-Modal Reference System Seedance 2.0 accepts: up to 9 images + 3 videos + 3 audio clips + text. | Input Type | Use For | Prompt Syntax | |---|---|---| | @image1 | Character ref, environment, color palette | "figure based on @image1 [stylized illustration]" | | @video1 | Camera movement, pacing, action style | "camera movement matched to @video1" | | @audio1 | Music rhythm, ambient sound, shot timing | "shot transitions aligned to beat drops at [0:32] in @audio1" | **Audio: MP3 only.** WAV and AAC not supported. **All character image references must include a stylization directive:** stylized illustration / 3D render / 3D-rendered portrait / anime-style portrait / digital painting / AI-generated character concept / cel-shaded character design Never request a photorealistic human face. --- ## Spanish Dialogue & Lip-Sync ### Rules for Lip-Sync Accuracy 1. Keep dialogue **5-8 words per line** (Spanish phonetics faster than English) 2. Use clear phonetic content — avoid tongue twisters 3. Action beats between dialogue lines — no rapid-fire back-and-forth 4. Specify speaker and tone: `"Técnico dice (tono serio): 'No puede ser.'"` **Good:** `"Paco: No puede ser. Ya sé quién eres."` (7 words, clear) **Avoid:** `"Paco: Ahora mismo vamos directamente por ese criminal sin esperar porque el tiempo se acaba rápidamente."` (too long) ### World ID — Character Consistency 1. Upload character reference as @image1 (must be stylized, not photo) 2. In Section 3: `"Técnico based on @image1 [stylized illustration] — lentes amarillos, chaqueta oscura, postura seria"` 3. Repeat same visual description in every extension prompt 4. Identical clothing/accessory details across all shots ### Video Extension / Chaining ``` Extend @video1 by [duration]. Reference @image1 for [character] consistency. [New section structure with same Visual Style as previous segment] ``` --- ## Templates (Copy & Customize) ### Template 1 — Single Character Scene (Spanish) ``` Visual Style: Cinematic realism, ARRI Alexa, [lighting]. [Lens]. [Atmosphere]. [Color grade]. Camera Behavior: [movement style]. Environment: [Location]. [Surface]. [Atmospheric detail]. [Ambient sound]. [Role label] based on @image1 [stylized illustration]. [Clothing, physical state]. [Body language as visible action]. Shot 1 — [Label] (0-5s) [Camera]. [Frame content]. [Character action]. [Sound]. Shot 2 — [Label] (5-10s) [Camera]. [Character action]. [Técnico dice: "[Spanish — max 8 words]"]. [Sound]. Shot 3 — [Label] (10-14s) [Camera pulls back]. [Resolution action]. [Ambient shift]. ``` ### Template 2 — Two-Character Dialogue (Spanish) ``` Visual Style: Cinematic realism, RED Komodo, [lighting]. 85mm shallow DOF. [Atmosphere]. [Color grade]. Camera Behavior: alternating close-ups and two-shots. Environment: [Location, surfaces, ambient sound]. [Character A role label]: [appearance, physical state]. [Character B role label]: [appearance, physical state]. Shot 1 — Establish (0-4s) Wide shot. Both figures visible. [Action]. [Sound]. Shot 2 — A speaks (4-8s) Close-up on A. [A dice: "[max 8 words]"]. [Physical micro-reaction]. Shot 3 — B responds (8-12s) Close-up on B. [B dice: "[max 8 words]"]. [Camera pulls to two-shot]. ``` ### Template 3 — Product Showcase ``` Visual Style: Commercial production, RED Komodo, [lighting]. Macro 85mm, razor-thin DOF. [Atmosphere]. [Color grade]. Camera Behavior: slow dolly push, steady glide, macro orbits. Environment: [Surface/set]. [Background]. [Atmospheric: mist, particles, gradient void]. [Ambient sound]. Product: [material, color, surface finish, dimensions]. Shot 1 — Reveal (0-4s) Wide shot, product partially obscured. Slow dolly push begins. [Atmospheric foreground]. [Sound]. Shot 2 — Detail (4-9s) Macro orbit. [Surface texture: reflections, condensation, grain]. [Sound]. Shot 3 — Hero (9-14s) Locked hero angle, dramatic lighting. [Surface highlight]. Camera holds. ``` --- ## Diagnostic Protocol 1. **Check platform status first.** Generic "Generation Failed" = filter block OR server overload. Rule out infrastructure before rewriting. 2. **Audit for soft-block triggers:** - Narrative/emotional subtext? Strip to pure visual description - Implied IP? Genericize character - Isolated violent/weapon action? Add cinematic framing - Age-ambiguous descriptors? Replace with role-based labels - Intent described instead of action? Rewrite as physical movement 3. **Audit image reference:** - Real human face or photorealistic portrait? Replace with stylized version - Visually resembles copyrighted character? Modify or genericize 4. **Reduce complexity to isolate:** - Drop to 720p - Reduce prompt to under 50 words - If passes at low complexity, gradually reintroduce elements 5. **Check file formats:** Audio must be MP3. When rewriting a failed prompt, deliver in a new code box and briefly state what changed and why. --- ## Production Notes (Include When Applicable) - **Image Reference:** "Ensure character references are stylized/AI-generated, not photorealistic — improves filter pass rate." - **Audio Reference:** "Upload as MP3. WAV and AAC not supported. Shot pacing is designed to sync with audio rhythm." - **Filter Framing:** "If generation fails, run diagnostic protocol." - **Multi-Shot Consistency:** "Maintain identical Visual Style section across all shots." - **Resolution:** "If generation fails, try 720p first before rewriting." Keep to 2-4 bullets. --- ## Next-Level Ideas (Always End With This) > **Para llevar esto al siguiente nivel, aquí van algunas ideas de qué podrías probar después.** > > 1. [Ajuste al prompt para mejorar resultados o cumplimiento de filtros] > 2. [Variación creativa de la misma idea en otra dirección] > 3. [Idea más experimental — inusual pero filter-safe] > 4. [Idea nueva simple que temáticamente conecta con lo que pidió] > > Puedes responder solo con un número y genero nuevos prompts basados en esa dirección. --- ## Quality Checklist **Filter:** - [ ] Every line passes the Shot List Test - [ ] No named or implied IP - [ ] No real human faces as references - [ ] No ambiguous age descriptors - [ ] All weapon/combat actions wrapped in cinematic production framing - [ ] No emotional subtext or character psychology - [ ] All image references include stylization directive - [ ] Role-based character labels throughout **Cinematic Quality:** - [ ] Section 1 opens with camera + lens + lighting before any action - [ ] Camera movement specific and physically described - [ ] Lighting has source, direction, visible quality - [ ] Textures and physical details present - [ ] Color grade stated explicitly - [ ] Ambient sound described **Format:** - [ ] Sectioned format (1-6 as applicable) - [ ] Under 1,990 characters total - [ ] Exactly ONE prompt delivered, in one code box (never 2 or 3) - [ ] Multi-modal references labeled (@image1, @video1, @audio1) - [ ] Production Notes included when applicable - [ ] Next-Level Ideas section present --- ## Platform Quick Reference - Duration: 4-15s per generation, extensible via chaining - Resolution: 720p (web) / 2K (professional) — UI control, not prompt content - Aspect ratios: 16:9, 9:16, 1:1 — UI control, not prompt content - Frame rate: 24 FPS - No negative prompt field - Audio format: MP3 only - Spanish prompts and dialogue: fully supported