Prompt style
Director brief
Concrete motion and visual intent.
Prompt lab
Seedance 2.5 Prompts
Build video directions around subject, action, camera, light, timing, sound, and constraints. Every example is a reusable creative template, not a fabricated model result.
Examples
6 feature templates
30s, references, editing, languages, and more.
Output claims
None
Templates are not represented as results.
Copy action
Browser only
Prompt text is not uploaded.
Prompt anatomy
Write in layers
Start with the non-negotiable subject and action. Add the location and time, then choose a coherent camera move, lighting design, pacing, and sound. For a 30-second sequence, divide the brief into a few time blocks with one clear transition between each.
Specific relationships are more useful than adjective piles: say how the camera moves relative to the subject, how light changes across the shot, and which details must remain fixed.
Subject + action
Who or what is present, and what changes during the shot.
Camera
Framing, lens feel, movement, speed, and where the move ends.
Light + atmosphere
Direction, color, softness, weather, particles, and reflections.
Sound + constraints
Dialogue, ambience, effects, continuity, exclusions, and final frame.
Reference control
Give every reference one job
The published Seedance 2.5 workflow supports richer combinations of images, videos, audio, scripts, and other references. A reference still does not explain its own role, so state whether it defines identity, product geometry, wardrobe, composition, movement, timing, color, or only mood.
- Name the exact attributes that must remain unchanged.
- Separate identity references from style references.
- Number references and assign each one a clear role instead of asking for a vague blend.
- Describe the intended first frame, camera path, and final frame.
- Avoid requesting a redesign when preservation is the priority.
Editing and speech
Constrain the change
For a precise edit, describe the smallest target region and explicitly protect timing, motion, lighting, audio, and every unaffected object. For multilingual dialogue, write the exact spoken lines, language order, pauses, performance, lip-sync goal, and whether subtitles or music are allowed.
Copy-ready
Feature-aware prompt library
Use these as starting points for the published 30-second, multimodal-reference, editing, and multilingual workflows. They remain writing templates, not claimed outputs.
Precision watch macro
Extreme macro of a mechanical watch assembling itself on brushed black steel. Begin on a single rotating gear, then pull back as bridges, hands, and sapphire glass lock into place. Crisp controlled motion, narrow depth of field, cool rim light, tiny metallic impacts, final top-down hero composition, no text.
Wind tunnel editorial
A model in sculptural silver fabric walks through a vast concrete wind tunnel. Side-tracking 50mm lens, fabric reacting naturally to changing air pressure, hard white light cutting through atmospheric haze, confident slow pacing, subtle industrial room tone, consistent face and garment.
The paper moon journey
Create one continuous 30-second story. 0–8s: inside a child’s dark bedroom, a handmade paper moon lifts from the desk as the camera slowly pushes in. 8–18s: follow the moon through drawings that animate into a luminous paper city, preserving the same moon shape and warm lamp color. 18–30s: the city folds back into the bedroom and the moon settles above the sleeping child. Tactile paper texture, seamless transitions, quiet wonder, soft pencil sounds, no text.
Assign every reference
Use [Image 1] only for the exact product shape, materials, color, and label; use [Image 2] only for the warm coastal location and sunset palette; use [Video 1] only for the slow circular camera path; use [Audio 1] as timing reference for cuts at 0s, 6s, 14s, and 24s. Preserve the product design in every shot. Build a 30-second premium launch film with no added logo, no redesign, and no unrelated visual style.
Change one region
Edit [Video 1]: replace only the red suitcase beside the main character with a matte black travel case. Preserve the character, face, wardrobe, hand contact, walking motion, camera movement, lighting, shadows, reflections, background people, original audio, duration, and frame timing. Do not regenerate or restyle any other region.
Four-language greeting
A presenter addresses the camera in one continuous medium close-up and says four short greetings in sequence: English “Welcome”, Chinese “欢迎”, Spanish “Bienvenidos”, and Japanese “ようこそ”. Keep the same identity, wardrobe, framing, room tone, and speaking volume. Match lip movement and expression to each language, leave a natural one-second pause between lines, no subtitles, no background music, clean studio audio.
Prompt FAQ
Better directions
What should a Seedance 2.5 prompt include?
Describe subject, action, environment, camera, lighting, sound, continuity requirements, and exclusions. For a 30-second sequence, add explicit time blocks and state what must stay consistent between them.
Are these confirmed Seedance 2.5 outputs?
No. They are creative templates prepared for testing. The page will label any future output with its verified model, settings, date, and source.
How do I use an image reference?
Assign each image, video, or audio reference one explicit role. Separate product identity, character identity, location, camera motion, timing, and style instead of asking the model to blend every reference equally.
Should I write one long paragraph?
A compact structured paragraph usually works better than keyword stuffing. Use concrete film language and avoid mutually conflicting lenses, lighting directions, or camera movements.
How should I write a video-editing prompt?
Name the exact region or object to change, describe the replacement, then list everything that must remain untouched: identity, timing, motion, camera, light, shadows, background, and audio.
Prompt update
Get future verified prompt tests.
Consumer access is live. Future on-site examples will still include the tested product surface, model version, settings, and provenance instead of anonymous showcase claims.