How to Use Seedance 2.5
A practical guide to 30-second generation, multimodal reference mapping, timestamped shot direction, storyboard control, and prompts that stay coherent to the final frame.
The five Seedance 2.5 capabilities that shape the workflow
Treat Seedance 2.5 like a directing system: references define the ingredients, timestamps define the sequence, and negative instructions define the boundaries.
4–30 second native clips
Plan a complete beat in one generation instead of stitching several short clips together. Longer clips work best when the prompt is split into 4–6 timed sections.
Up to 50 multimodal references
Combine up to 30 images, 10 videos, and 10 audio files. Assign every asset a specific job: identity, environment, motion, camera, style, prop, voice, or rhythm.
More natural human footage
Improved lighting, camera motion, expression, performance, and movement continuity help generated people feel closer to photographed footage.
Second-by-second direction
Use continuous timestamp ranges to tell the model what should happen, how the camera should move, and what should be heard during each part of the clip.
Storyboard-led generation
Map storyboard panels to time ranges so the model understands narrative order and transitions.
A production workflow for 30-second generations
Make every instruction easy to assign, time, and verify instead of merely making the prompt longer.
Give every image, video, and audio file one clear job
The model accepts up to 30 images, 10 videos, and 10 audio files, with no more than 50 references in total. Each file should add unique control.
Images
Up to 30 images, including high-resolution references.
- Use clean portraits for character identity.
- Use separate images for environment, product, prop, or style.
- Label storyboard order and time ranges.
Video
Up to 10 clips with a combined input duration of 30 seconds.
- Trim each clip to the exact motion you need.
- State whether it controls action, framing, pacing, or camera language.
- Do not assign unrelated roles to one clip.
Audio
Up to 10 files with a combined input duration of 30 seconds.
- Identify voice, dialogue, ambience, effect, or rhythm.
- Describe when each sound enters.
- Explicitly forbid music when it is not wanted.
See the workflow in actual Seedance 2.5 outputs
Watch finished videos demonstrating longer continuity, human realism, multimodal control, and timestamped direction.
30-second narrative continuity
A complete sequence that keeps the subject, environment, and visual direction connected.
Natural live-action detail
A verified Seedance 2.5 output for judging skin, hands, material texture, lighting, and motion.
Multimodal reference control
An output built from environment, character, style, and camera references with explicit roles.
Timestamp-led direction
Evaluate whether actions, camera instructions, and transitions arrive in the intended order.
A timestamped Seedance 2.5 prompt structure
Use continuous time ranges, one main action, and one primary camera move per range.
FORMAT [Duration], [aspect ratio], [visual treatment], [single shot or edited sequence]. REFERENCE MAP Character: use [image reference] for identity and wardrobe only. Environment: use [image reference] for location and lighting. Motion/camera: use [video reference] for [specific movement] only. Audio: use [audio reference] for [voice / rhythm / ambience]. Do not copy [unwanted element] from the references. TIMELINE 0–5s: [subject action]. Camera: [one move]. Audio: [sound cue]. 5–12s: [next action]. Camera: [one move]. Transition: [continuity instruction]. 12–20s: [next action]. Camera: [one move]. Audio: [sound cue]. 20–26s: [payoff]. Camera: [one move]. 26–30s: [ending image and hold]. CONSISTENCY Keep [identity / product / wardrobe / lighting / location] consistent throughout. NEGATIVE INSTRUCTIONS No [extra people / subtitles / logos / background music / abrupt cuts / camera shake].
A 30-second multimodal product-film prompt
Reference roles, timed shots, continuity rules, and negative instructions are separated.
30-second 16:9 cinematic product film, realistic live-action look, five-shot sequence. Reference map: use Image 1 for the woman’s identity and cream suit only. Use Image 2 for the perfume bottle shape, label, and amber glass. Use Video 1 only for the slow tabletop orbit. Use Audio 1 for the soft spoken line at 18 seconds. Do not copy the room or wardrobe from Video 1. 0–5s: Dawn light crosses a quiet stone dressing table. The perfume bottle stands centered in a thin layer of mist. Camera: locked macro close-up with a very slow push-in. Audio: room tone only. 5–11s: The woman enters frame and reaches toward the bottle without touching it. Camera: medium side profile, gentle left-to-right dolly. Keep her face and cream suit consistent with Image 1. 11–18s: Her fingers lift the bottle; warm reflections move naturally through the amber glass. Camera: use Video 1’s slow orbit. No change to the label or bottle proportions. 18–24s: She looks toward the window and says the line from Audio 1 once, with natural lip sync. Camera: steady medium close-up. No music. 24–30s: Cut back to the bottle as sunlight fills the frame. Camera: slow pull-back, then hold the final composition for two seconds. Keep the woman, product, room, and sunrise lighting continuous. No extra people, subtitles, added logo, background music, or handheld shake.
Fix the instruction that caused the failure
Diagnose identity, reference roles, timeline, camera, and sound separately before rewriting everything.
The second half drifts
Reduce events, clarify time ranges, and restate the subject or location at the transition that fails.
References fight each other
Remove redundant assets and give every file one responsibility plus explicit exclusions.
People duplicate or change
Map each character to one identity reference, keep names consistent, and forbid additional people.
Camera motion looks unstable
Use one primary move per block and specify a pan, dolly, orbit, tracking shot, or locked frame.
Unwanted subtitles or music appear
Add direct negative instructions: no subtitles, logos, or background music.
The storyboard order is ignored
Number panels, map continuous time ranges, and describe every transition.
Seedance 2.5 tutorial questions
How should I structure a 30-second Seedance 2.5 prompt?
Use 4-6 continuous timestamp ranges. Give each range one main action, one primary camera move, and any audio cue that must happen there. End with continuity and negative instructions.
Do I need to use all 50 reference slots?
No. Use only references that add unique control. Seedance 2.5 can accept up to 30 images, 10 videos, and 10 audio files, but redundant or conflicting references make the result harder to direct.
How do I use a storyboard with Seedance 2.5?
Number the panels, map each panel to a time range, describe the action and camera for that range, and explain how the scene should transition into the next panel.
Turn the timeline into a video
Start with the template, assign every reference a role, and generate your first Seedance 2.5 sequence.