Use timestamp prompts when the order of actions matters. Start with the subject and setting, divide the clip into a small number of time ranges such as 0–3s, 3–6s, and 6–10s, and give each range one primary action plus any camera instruction that must happen there. Timestamp labels can organize a request, but they do not guarantee exact frame-level timing. Check the generated clip and adjust one beat at a time.
Why timestamp prompts help
A broad prompt may describe the right subject but leave the order of events open. A timestamp structure gives the model a readable sequence: establish the subject and location; assign an action to each time range; describe camera movement where it belongs; and reserve the final beat for a pose, product view, or clean ending. A longer paragraph can still contain competing actions. The useful unit is a bounded beat with a clear change from the previous beat.
For a current overview of the Seedance 2.5 workflow, see the Seedance 2.5 model page. For prompt-size and conflict checks, also see how to fit a clear Seedance 2.5 prompt.
A practical timestamp structure
| Block | What to specify | Why it matters |
|---|---|---|
| Subject anchor | Person, product, or object; visible identity details | Prevents the subject from changing between beats |
| Scene | Location, surface, time of day, and background | Gives every timestamp the same world |
| Action beats | One main action per time range | Reduces simultaneous or skipped actions |
| Camera | Framing and one movement at a time | Keeps camera direction from fighting the action |
| Ending | Hold, look, product pose, or other final state | Gives the clip a reviewable endpoint |
Keep the number of beats proportional to the duration. A 4–5 second test usually needs fewer beats than a 15–30 second narrative.
Illustrative 10-second example
This is an illustrative prompt, not a tested result. It assumes one reference image of a person and does not claim that the output will match these timings exactly.
Create a 10-second vertical video of the same adult cyclist from @Image1, wearing the same navy windbreaker and matte black helmet, on a quiet riverside path at early morning. Keep the cyclist face, clothing colors, bicycle shape, and background direction consistent. Timeline: 0–3s wide establishing view; cyclist stationary beside the bicycle and looking toward the path, camera steady. 3–7s cyclist starts riding forward at a relaxed pace; one smooth side-tracking move, no orbit or zoom. 7–10s cyclist slows near the river, turns the front wheel slightly toward camera, and holds a natural finishing pose. End on a stable medium-wide frame. Use soft morning light, realistic bicycle motion, natural body balance, and one continuous shot. No cuts, no new people, no extra bicycles, no text, no logo, and no sudden change of location.
The reference label is part of the example. In a real generation, attach the image again and use the asset label shown by the current interface.
How to revise a weak result
If the subject changes, shorten the description and repeat only identity details that matter. Add a stable reference image when supported. If an action is skipped, remove a competing action from that time range. Do not ask the subject to start riding, wave, turn toward camera, and pick up a bag in the same beat. If the camera instruction conflicts with the action, choose one movement that supports the beat. If the clip feels unfinished, describe the final state: hold the pose, face the product toward camera, stop the camera, or leave a clean final frame. For recurring failures, see /blog/seedance-2-5-prompt-troubleshooting and /blog/seedance-2-5-prompt-examples.
Timestamp prompts versus longer videos
Timestamp labels do not change the selected duration or credits. Choose the model, aspect ratio, resolution, and duration in the generator, then review the current estimate before submitting. The public product page currently describes Seedance 2.5 as supporting 4–30 second outputs and up to 50 reference materials. Treat that as the current page description, not a promise that every mode or account exposes every setting. The generator estimate is deciding evidence. See /blog/seedance-2-5-credit-preview.
FAQ
Do timestamp prompts guarantee exact second-by-second cuts?
No. They communicate intended timing and order, not exact frame boundaries. Inspect the result and simplify the prompt when a transition is missed.
How many timestamps should I use?
Use the fewest beats that answer the creative question. Two or three ranges are a practical starting point for a short test.
Should every timestamp include a camera move?
No. Keep the camera steady when the action is the priority.
Can I use the same prompt for every aspect ratio?
Not reliably. Recheck framing, subject visibility, and the final pose after changing between landscape, portrait, or square output.
Summary
Timestamp prompts work best as a reviewable sequence. Anchor the subject, keep one primary action per range, separate camera changes from story changes, and define the final state.
