01
Choose the workflow that matches your starting material
Start with Text to Video when the scene can be invented from a written brief. Use Image to Video when one still already defines the subject and opening composition. Choose References to Video when identity, products, wardrobe, environment, or style are distributed across several assets. Use Video to Video when an existing clip already contains useful motion, timing, framing, or performance.
- Text to Video: invent a complete scene from written direction.
- Image to Video: animate one visual anchor while protecting its important details.
- References to Video: combine several sources by assigning every image a clear role.
- Video to Video: preserve useful source motion while redesigning appearance or atmosphere.
02
Build the prompt as a sequence, not a list of adjectives
Start with the subject and environment, then describe the action in chronological order. Add camera direction, continuity constraints, visual treatment, sound, and a defined ending. This gives a longer Seedance 2.5 request a beginning, development, and final beat instead of several disconnected visual ideas.
- Subject and setting: identify who or what is on screen and where the shot takes place.
- Ordered action: explain the opening action, the next beat, and how the sequence resolves.
- Camera: state framing, angle, movement, pace, and any transition between views.
- Continuity: name the identity, wardrobe, product, prop, or environment details that cannot drift.
- Picture and sound: connect lighting, ambience, dialogue, music, and effects to specific visual moments.
- Final frame: describe the pose, composition, reveal, or emotional beat where the shot should end.
03
Use Text to Video when the scene can be invented
Use Text to Video for cinematic concepts, environmental shots, social hooks, animation, visual experiments, and product ideas that do not require exact reference matching. Describe the result as a shot to direct rather than a topic to illustrate.
04
Let one image define appearance and composition
The image should clearly show the subject, composition, perspective, environment, and details that matter. A clean source frame gives the prompt more room to direct motion instead of repairing unclear visual information.
- Use sufficient resolution and avoid severe compression or motion blur.
- Keep important faces, products, hands, and logos readable in the starting frame.
- Choose a composition with enough physical space for the intended movement.
- Match the source aspect ratio to the planned output whenever possible.
05
Give every reference one clear responsibility
Do not ask the model to infer why each image was uploaded. Name each reference and describe what it controls. Avoid giving two conflicting images responsibility for the same face, garment, product, or environment.
- Image 1 defines the character’s facial identity and hairstyle.
- Image 2 defines wardrobe, jewelry, and color details.
- Image 3 defines product shape, materials, logo placement, and proportions.
- Image 4 defines the room layout and architectural background.
- Image 5 guides lighting contrast, color palette, and atmosphere only.
06
Separate preserved motion from the intended transformation
A source video already defines time, performance, blocking, and camera behavior. State which of those elements must remain stable, then describe the new environment, material, lighting, weather, atmosphere, or visual treatment. Avoid requesting a new action that contradicts the source movement.
- Preserve the timing, subject position, framing, and useful performance beats.
- Describe visual changes with observable materials, lighting, setting, and atmosphere.
- Choose footage whose movement already supports the intended transformed result.
- Use Text to Video when the desired action is fundamentally different from the source clip.
07
Write action and camera direction separately
Subject movement and camera movement are different instructions. Describe what the subject does first, then explain how the camera responds. A clear separation helps prevent a prompt from asking for several incompatible movements at once.
Use concrete verbs such as turns, reaches, opens, accelerates, pauses, or looks toward the window. For the camera, use a deliberate path such as a slow dolly in, a lateral tracking move, a controlled orbit, or a locked medium shot. Keep the number of major actions realistic for the selected duration.
For a four- or five-second clip, one principal action and one camera move are usually easier to read. A longer request can support connected beats, but each beat should still lead naturally into the next.
08
Protect the details that define the shot
Continuity instructions should be selective. Protect the details that carry identity or brand value rather than asking the model to freeze every pixel. State exactly what must stay stable as movement develops.
- Preserve facial identity, hairstyle, makeup, and defining accessories.
- Keep wardrobe colors, materials, logos, and product geometry unchanged.
- Maintain the location layout, direction of light, and spatial relationship between subjects.
- Avoid duplicate limbs, sudden costume changes, unwanted text, or objects appearing between beats.
09
Place sound cues beside the visual event they support
When native audio is enabled, describe dialogue, ambience, music, and effects at the visual beat where they occur. A sound instruction is more useful when it explains timing and purpose instead of simply requesting “cinematic audio.”
Practical examples
Prompts you can adapt to your own brief
Replace the subject, environment, action, protected details, duration, format, and sound direction. Keep the structural relationship between those decisions intact.
Prompt example
Product reveal
A controlled commercial shot with a clear final composition.
Create an 8-second 16:9 product film of a brushed-metal wristwatch on a dark stone pedestal. Begin with a tight macro view of condensation moving across the bezel. The watch rotates slowly as a narrow warm light travels from left to right, revealing the engraved crown and keeping the dial markings, case geometry, and logo unchanged. The camera makes one smooth dolly out to a centered medium product shot. Add a low room tone, a subtle mechanical tick, and one restrained impact as the logo becomes fully visible. End on a stable hero frame with clean negative space on the left.
Why it works
It limits the shot to one product action and one camera path, protects brand-critical details, connects sound to the reveal, and defines the final layout.
Prompt example
Vertical social hook
One surprising action designed for a short 9:16 clip.
Create a 6-second 9:16 social video in a quiet laundromat at midnight. A woman in a yellow raincoat opens a washing machine and a dense cloud of tiny glowing butterflies pours out, briefly circling her before filling the ceiling. Begin in a locked medium shot, then make a quick controlled dolly in as the first butterflies appear. Keep her face, raincoat, and the machine geometry stable. Use fluorescent room tone, a soft mechanical click, and a rising shimmer as the butterflies emerge. End with her looking upward while the room glows gold.
Why it works
The clip has one visual hook, one camera response, a compact sound progression, and a final frame that completes the idea.
Prompt example
Portrait animation
Subtle performance that preserves a carefully composed portrait.
Animate this portrait into an 8-second 9:16 cinematic shot. The subject takes a quiet breath, looks from the window toward camera, and gives a restrained half-smile. Her hair and the sheer curtain move gently in the same breeze. Make a very slow dolly in while preserving her facial identity, freckles, hairstyle, earrings, blue linen shirt, window geometry, and soft side lighting. Add quiet room tone, distant city rain, and a subtle fabric rustle. End in a stable close-up with her eyes meeting camera.
Why it works
The source image continues to control appearance and composition while the prompt adds achievable movement, one camera path, protected details, sound, and an ending.
Prompt example
Fashion performance with protected identity
Five references coordinate identity, wardrobe, location, and lighting.
Create a 12-second 16:9 fashion film in a restored railway hall. Image 1 defines the performer’s face and short black hairstyle. Image 2 defines the cream structured coat, gloves, and silver earrings. Image 3 defines the exact red travel case and brass hardware. Image 4 defines the station architecture and floor pattern. Image 5 guides only the warm backlight and muted green palette. The performer walks toward camera, places the case down, then turns as a train light passes behind her. Track backward at chest height in one smooth move. Preserve facial identity, coat construction, case geometry, and station layout. Add distant rail ambience and a restrained orchestral rise. End on a stable three-quarter portrait beside the case.
Why it works
Every image has a non-overlapping role, the protected details are named again, and the motion remains simple enough to preserve continuity.
Prompt example
Video transformation with protected motion
An existing performance supplies timing while the prompt redesigns the visual world.
Transform the uploaded 8-second dance clip into a moonlit performance inside a shallow mirrored pool. Preserve the performer’s exact timing, body movement, camera path, framing, and final pose. Replace the studio with a dark open courtyard, change the wardrobe surface to reflective silver fabric while keeping its silhouette, and add cool moonlight with warm lantern reflections moving across the water. Keep the performer’s identity stable and prevent changes to limb count or choreography. Add restrained night ambience, water movement synchronized with each step, and one low tonal impact on the final pose.
Why it works
It clearly separates protected source motion from the new wardrobe material, environment, lighting, atmosphere, and sound treatment.
Final review
Check the brief before spending credits
- Did you choose the simplest workflow for the available source material?
- Can the main action fit within the selected duration?
- Is the camera path compatible with the subject movement?
- Does every image or video source have a clear responsibility?
- Are identity, product, wardrobe, and environment details protected?
- Are visual conflicts resolved with an explicit priority?
- Are sound cues tied to visible moments?
- Does the brief define a final frame or ending beat?
Good to know
Frequently asked questions
How long should a Seedance 2.5 prompt be?+
Use enough detail to define the complete shot without repeating the same visual idea. A concise paragraph can direct a simple clip, while a longer sequence may need separate sentences for action, camera, continuity, sound, and the ending.
Should I include every camera term I know?+
No. Choose one primary framing and one coherent camera path for a short clip. Several incompatible camera moves can weaken the visual hierarchy and make the result less predictable.
Can I reuse these prompts exactly?+
Yes, as a starting point. Replace the subject, environment, action, protected details, output format, and sound cues so the prompt matches your assets and production goal.
Which Seedance 2.5 workflow should I use?+
Use Text to Video for a scene invented from words, Image to Video for one visual starting frame, References to Video when several assets define different visual responsibilities, and Video to Video when existing footage already provides useful motion and timing.
Should I use as many references as possible?+
No. Use the smallest coherent set that communicates the required visual facts. Every additional reference should add a distinct role without contradicting the others.
How do I reduce unwanted visual changes?+
Keep the action achievable, use a coherent camera path, explicitly protect important identity and product details, resolve source conflicts, and describe a stable ending frame.