SEEDANCE 2.5 PROMPT GUIDE
Prompt adherence tips, templates, workflows and prompts for Seedance 2.5. Generate amazing AI videos.
Dreamina
Seedance 2.5
Prompt Guide
Stop guessing. Every formula, syntax rule, and control structure for text-to-video, multi-reference creation, 30-second stories, editing, and extension — in one guide.
One model. Four jobs.
Text to video
State what you want. Subject and action carry the whole shot; everything else is optional refinement.
Reference-driven
Up to 50 image, video, and audio references — each with a role you write out explicitly.
Editing
Change a subject, a background, a light, or one audio layer inside an existing video.
Extension
Generate before or after a source clip, matched to the boundary frame.
Basic
Prompting
The core formula, reference roles, and the special syntax for music, sound effects, dialogue, and subtitles.
Six blocks. Two required.
Subject
Who or what.
Action / Event
What happens. The foundation of the video.
Scene
Location, time, weather, spatial layout.
Visual style
Lighting, color, material, texture, mood.
Camera
Shot size, angle, movement, focus, cuts.
Audio
Dialogue, voice, ambience, SFX, music.
The formula, filled in.
A ceramic artist finishes a pale blue cup in a studio at dawn, lifts it from the wheel, and places it in the center of a wooden shelf.
Soft morning light enters through the window. The wet clay has a delicate sheen, and the workbench remains tidy.
Begin with a medium shot of the wheel-throwing process, slowly push in toward the cup's surface texture, then cut to a frontal view of the shelf.
Retain the low hum of the pottery wheel, the friction of clay, and subtle indoor ambience.
Up to 50 refs. Fewer is steadier.
Images
Max 30 images, each no larger than 4K.
Recommended: 1–8 distinct subjects across subject-reference images.
Videos
Max 10 videos, 30 seconds combined.
Recommended: 1–5 subjects, 5–10 seconds per subject video.
Audio
Max 10 clips, 30 seconds combined.
Keep only dialogue, voice, ambience, or music relevant to the task.
Editing
One source video plus reference images.
Recommended: source under 20 seconds, 1–5 reference images.
Say what to use. And what not to.
Never rely on labels inside the image, and never make the model guess which reference belongs to which subject.
Defines the ceramic artist's facial features, hairstyle, and dark green apron.
Do not use the image background.
Defines the wooden workbench, window placement, and morning light.
Do not use the people in the image.
Defines the pacing of throwing clay, lifting the cup, and placing it down.
Do not use the person's identity, clothing, or scene.
Four brackets. Four channels.
Music
(Soft, rhythmic piano music plays in the background)
Sound FX
<A bell rings in the distance>
Dialogue
{Hello, welcome back.}
Subtitles
【Chapter One: Departure】
New in 2.5
Multi-reference creation, 30-second stories built from stages, parameter locking, editing, and extension.
Don't list materials. Define relationships.
Name & map each subject
One line per person, prop, and scene. Bind it to one material and state which attributes to take.
Group by type
[Characters] [Props] [Scenes] [Motion & Audio]. Add a no-swap rule inside each group.
Profile key subjects
For anyone crossing scenes: appearance, fixed prop, locations, motion refs, and a "do not use" list.
Select refs per scene
Each scene lists only the materials it uses, its event, and its end state.
Groups, profiles, scene picks.
One stage. One change.
Every stage gets one primary change, plus an explicit, visible end state the next stage continues from.
Initial state: florist behind the workbench; stems, scissors, and wrapping paper on the tabletop.
Primary event: florist arranges the stems and trims them to length.
End state: bouquet in the left hand, scissors back on the right side of the bench.
Continue from previous: same identities and clothing; florist still holds the bouquet.
Primary event: assistant unfolds the paper; florist places the bouquet in and ties a green ribbon.
End state: wrapped bouquet flat in the center of the bench, bow facing camera.
Closing beat.
Primary event: assistant picks up the bouquet and places it on the pickup shelf.
End state: bouquet centered on the shelf; both characters behind the bench inspecting it.
Budgets, not edit points.
0–3s … 3–7s … 7–12s
Allocate pacing across consecutive, non-overlapping windows.
At 5 seconds, whip-pan left
Pin a single key event — a transition, entrance, or beat.
3s after the button press…
Describe a delay between two events instead of absolute time.
Some settings lock themselves.
| Task type | Aspect ratio | Duration |
|---|---|---|
| Video editing | Preserves the input video's ratio. Cannot be set. | Approximately preserves input duration; up to ~0.3s difference. Cannot be set. |
| First / last frame | Uses the first image's ratio. Match both images or the last frame stretches. | Can be set. |
| Video extension | Preserves the input video's ratio. Cannot be set. | Can be set. |
Five blocks. One master.
Add, remove, replace, or adjust an object, region, or audio category — whole video or one time range.
@Video 1 is the sole editing master: scene, actions, camera, occlusion, audio, event order.
The reference defines the target's named attributes — nothing else from it.
Modify only this object, region, range, or audio layer — and state the target count.
Everything from @Video 1 that must not change: motion, audio, timing.
Swap it. Replace it. Mute it.
State the exact target count in the video, then hand over the timeline.
[Timeline Inheritance] the target inherits every appearance, motion, occlusion, and exit of the original — timing, duration, path, speed.
Take only spatial layout, materials, depth of field, ambient color, and lighting direction from the reference.
Modify only outside the subject's silhouette. Identity, features, hair, clothing, expression, position, size, motion: unchanged.
Dialogue, language, voice, music, and SFX are separately editable layers.
"Remove only the original background music. Keep dialogue, lip sync, ambience, and action sound effects."
Align the boundary before you write.
Describe what happens first, then define the source's first frame as the explicit end state of your segment — pose, prop position, spatial relationships, composition, lighting, motion direction.
Name anything that must not appear early: later characters, props, and effects belong after the source begins.
"Then connect to the source video" alone lets characters arrive too early or the image drift after arrival.
Describe the continuous state of the source's last frame first — subject pose and orientation, prop position, background, camera and composition, lighting, motion direction. Then say what happens next.
Hold identity, clothing, key props, background layout, and axis of action across the whole extension.
Keep each subject one continuous instance: never duplicated, split, or gaining parts.
Advanced Control
Keyframes, storyboards, blockouts, one-click video, seamless transitions, performance direction, and camera language.
One role per image.
Opening composition, position, pose, prop state, camera direction.
Visible end state of stage 1.
Visible end state of stage 2.
Ending composition, pose, prop state, camera direction.
Describe each anchor image separately — never "@Images 1 and 2 are the first and last frames."
First and last images must share an aspect ratio; the output locks to the first image.
Extra references supplement attributes only — they must never replace anchor compositions.
Inherit structure. Not the look.
State the reading order (left to right, top to bottom), then describe each panel's action, shot size, style, and audio.
≤15 panels · clean line art · minimal labels · never inherit the grid's style or text.
Simple geometry previewing action, paths, blocking, camera movement, and cuts.
Pin down: path (trajectory, direction, entrances) · camera (position, path, speed change) · light (direction, brightness, when it changes) · cuts (positions, composition either side) · audio (what to inherit, if anything).
Map every geometric object to its final subject: "the tall cylinder corresponds to <Guide>."
Complete modeling that needs new materials, colors, characters, scenes, or style.
Preserve structure, action, spatial layout, camera position, movement, and cuts. Re-render only the attributes you name.
Keep the source clean: no path lines, coordinate axes, controllers, or camera frustums.
Never write "make a video."
01 Material roles — what each image and the style video are for.
02 Arrangement — exact order, or permission to group by theme.
03 Image motion — parallax, push-in, or small local action; what stays stable.
04 Final style — editing rhythm, transitions, subtitles, color.
05 Audio — dialogue, ambience, sound effects, music.
A style-reference video contributes rhythm, transitions, stickers, and music only — never its identities or locations.
Before clip → Trigger action → Camera move → Transform → Arrival state → Audio blend
Dive / reverse: camera direction, speed change, when the next scene starts.
Rotation: pose, rotation direction, how clothing and background change.
Foreground occlusion: when the object fills frame, and what composition follows.
Object morph: corresponding shapes, materials, and the transformation.
Push / focus: movement, focus target, continuous spatial relationship.
A generated bridge is continuity — not a pixel-identical edit splice.
Don't name the feeling. Show it.
"The actor looks tense, then relieved."
Applause comes from behind the stage. The young actor's fingers stop on the program, the gaze turns slowly toward the curtain, the shoulders stay tense. After confirming the curtain call is over, the actor exhales softly, shoulders relax, a restrained smile appears, and the eyes slowly well — but the actor never turns to leave.
Say the term. Then say what it looks like.
extreme wide · wide · medium · close-up · extreme close-up
push in · pull out · pan · lateral · follow · orbit · dive · dolly out · tilt up · handheld shake
low angle · overhead view · first-person view
One-take: the subjects, spaces, and events the camera passes through, in order.
Dolly zoom: subject size to preserve; whether the background nears or recedes.
Aerial: viewing height, movement direction, area to reveal.
FPV: flight or traversal path, speed, turns.
Bullet time: the action to freeze or slow, and the orbit direction.
Handheld: the subject followed, and the amount of shake.
Bounce speed ramp: where it accelerates, decelerates, rebounds, and rests.
Run the checklist.
- Subject and primary action or event are stated clearly.
- Every material says what to use — and what not to use.
- Every character, product, and prop is named and bound to a reference.
- References are selected per scene, not forced to appear all at once.
- Each long-video stage has one primary change and a clear end state.
- Headcount, clothing, prop ownership, and spatial relationships stay consistent.
- Edits define the sole master, scope, target quantity, and content to preserve.
- Abstract emotions and camera terms are paired with observable cues.
- Anchor frames have one role each; first and last share an aspect ratio.
- Storyboards and blockouts state which structure to inherit — coarse or fine.
- Locked aspect-ratio and duration rules are respected for the task type.
- Extensions were checked at the boundary image, motion trend, and audio.
- One-click prompts define roles, order, motion amount, editing style, and audio.
- Transitions define both clips' roles, the trigger, the process, and the arrival state.
Name everything.
Bind everything.
Say what not to use.
Every failure mode in this guide comes from the same place: a material the model had to guess about. Precision costs a sentence. Guessing costs a render.
WANT TO JOIN THE LET'S AI SKOOL COMMUNITY
It just launched!
🔑 AI Video Mastery
🔑 Amazing AI Images
🔑 Prompt Engineering
🚀 Non-Stop Updates
🚀 Prompt Playbooks
🚀 Live Workshops
🎁 Plus Tons of Resources + Tons of Prompts Databases
Stop Guessing. Start Dominating.
The exact workflows, prompt secrets, and powerful frameworks behind the top AI platforms — with a community that levels up every time the tools do.
Join Let's AI — $19/Month → 👑 Why The Everything Bundle?
You'll never need prompts again!
That's because you'll have prompts for every category imaginable. From AI images and video, prompt generators, content, secret tokens and so much more. PLUS you'll have lifetime access and will receive non-stop updates for life. Today 15,000+ prompts, but in the future, 50,000+ prompts as it will continue to grow.
🧬 New databases/categories just added:
🎬 The AI Video Engine
🍌 Nano Banana Category
📚 The Content King
THE EVERYTHING BUNDLE
🚀 WHAT'S INSIDE?
✅ Over 10,000 prompts (and counting)
✅ Prompts Portal
✅ AI Video Engine
✅ AI Influencer Database
✅ Freepik Custom Database
✅ MEGA Prompts Database
✅ Leonardo AI database
✅ Prompt Generators database
✅ Photoreal Ebook
✅ Nano Banana Prompts🍌
✅ JSON Prompts
✅ Unique Keywords Database
✅ Content King Database
✅ Canva Prompts Website
✅ Lifetime Access
✅ Constant Updates
✅ Immaculate Organization
✅ Bonus Items