5 Seedance 2.5 Prompts With Real Video Examples
What do Seedance 2.5 prompts produce when you run them through PiAPI? These five original outputs cover text-to-video, image reference, first-and-last-frame control, motion transfer, and product relighting. Each example includes a recreation prompt, the supplied input where available, settings, the real output, and the most useful takeaway.
Developers, marketers, and AI video creators can use these examples as practical starting points. The recreation prompts describe the supplied inputs and observed outputs; they are not task-log transcripts or a controlled model benchmark, and another run may produce a different result.
Key takeaways
- Five generated videos covered three supported workflows at 720p.
- Clear subject, motion, camera, lighting, and reference instructions produced the most readable results.
- Reference media helped control transformations and camera rhythm.
- Actual output shape followed the supplied reference in cases where it differed from the planned aspect ratio.
- The five requests specified 26 output seconds, with a documented output-only cost of $15.60 at $0.60 per second.
On this page: Results · Method · Five prompts · Prompt lessons · API example · FAQ
What Makes a Good Seedance 2.5 Prompt?
Definition: A Seedance 2.5 prompt is a compact shot brief for PiAPI's video task. It names the subject, motion, camera behavior, visual treatment, and role of each reference. For a 4–15-second output, it should also state what must remain stable and what must not appear.
Which Seedance 2.5 Prompt Pattern Should You Use?
| Goal | Input pattern | PiAPI mode | Give the reference this role |
|---|---|---|---|
| Create an original scene | Text only | text_to_video | No reference; describe the complete shot |
| Animate a product or subject | One or more images | omni_reference | Preserve identity, shape, materials, or style |
| Control the start and destination | First and last images | first_last_frames | Define the two endpoint states |
| Reuse motion or camera pacing | Video | omni_reference | Guide movement without copying the original subject |
| Change a visual treatment | Image or video | omni_reference | Preserve the subject while changing lighting or setting |
Seedance 2.5 Video Examples at a Glance
All five outputs were generated at 720p. The live Seedance 2.5 API documentation lists 480p and 720p output, integer durations from 4 to 15 seconds, and support for image, video, and audio references.
| Example | Input pattern | Actual output | Documented cost basis | Main result |
|---|---|---|---|---|
| Rainy greenhouse tram | Text only | 5.08s, 1280×720 | $3.00 | Cinematic lateral tracking with strong warm/cool contrast |
| Watermelon product reveal | Image reference | 5.08s, 1280×720 | $3.00 | Stable product shot with mist and a moving light sweep |
| Desk lamp unfolding | First and last frames | 6.08s, 1280×720 | $3.60 | Smooth mechanical transformation with a stable camera |
| Delivery robot tracking | Example 1 as motion reference | 5.08s, 1280×720 | $3.00 output + about $1.52 video input | Camera rhythm transferred into a new subject and setting |
| Desk lamp relighting | Image reference | 5.08s, 1280×720 | $3.00 output | Clean cool-lit product treatment with stable motion |
The output-only total is $15.60. Using the documentation's half-rate formula for Example 4's 5.08-second video reference adds about $1.52, for an estimated set total of about $17.12 before account-specific billing adjustments.
Observed in this PiAPI example set: Five requests specified 26 output seconds, while the returned 720p MP4 files total about 26.42 seconds. The set covers text-only, image-reference, first/last-frame, and video-reference workflows and was reviewed on August 3, 2026; it is not a model benchmark.
How We Generated These Seedance 2.5 Examples
On PiAPI, Seedance 2.5 is accessed with model seedance and task type seedance-2.5. This is an output-led prompt guide rather than a task-log audit: the recreation prompts were written from the supplied inputs and observed outputs. The returned files are H.264 MP4 videos at 1280×720 and 24 frames per second.
The prompt plan set generated audio to off, but every returned file contains an AAC audio stream. The publication copies remove those audio streams while preserving the original downloads for task-record verification.
For broader model context, read the Seedance 2.5 model and playground guide. For endpoint, pricing, and limit details, use the Seedance 2.5 API specifications, with the live API documentation taking precedence if the two pages differ.
Example 1: Cinematic Text-to-Video
This baseline uses no reference media. The prompt separates the subject, camera direction, environmental motion, lighting, and exclusions.
Recreation prompt
At blue hour, a small yellow maintenance tram glides slowly through a rain-soaked glass greenhouse filled with tropical plants. The camera tracks parallel to the tram at waist height in one continuous shot. Water beads on the glass, the wheels rotate naturally, and nearby leaves shift gently from the tram's airflow. Warm amber light glows inside the tram against the cool blue greenhouse. Realistic cinematic lighting, stable geometry, natural motion, no cuts, no text, no people.Settings: Text to video · 5 seconds · 720p · 16:9 · audio off
What the result showed: The output captured the rainy greenhouse, parallel tracking composition, and warm tram light against a cool blue environment. The tram stayed stable, although foreground foliage hid most of the wheel movement requested in the prompt.
Example 2: Product Image-to-Video Prompt
This example uses an image reference to turn a static watermelon into a commercial product reveal. The prompt gives the reference one clear job: preserve the product while the model changes the presentation around it.
Recreation prompt
Use @image1 as the exact product reference. Preserve the product's silhouette, materials, colors, label placement, and proportions. Create a premium vertical product reveal: the product stands on a dark stone pedestal while a narrow warm light sweeps slowly across its surface. The camera makes a smooth 20-degree arc from the front-left to a centered close-up. Add subtle drifting mist behind the pedestal, never in front of the product. One continuous shot, clean commercial lighting, stable edges, readable packaging where visible in @image1, no added text, no extra products, no hands.Settings: Omni reference · 5 seconds · 720p · planned 9:16 · actual 16:9 · audio off
What the result showed: The watermelon remained clear and stable while the light sweep and mist created a polished product-shot treatment. The supplied reference is horizontal, and the output was also horizontal even though the recreation plan requested 9:16. That is consistent with the documented reference-aspect-ratio behavior.
Example 3: First-and-Last-Frame Transformation
First and last frames are useful when a scene needs a defined starting state and destination. Here, the same desk lamp begins folded and ends fully open.
Recreation prompt
Begin exactly from @image1 and end exactly on @image2. In one continuous locked-camera shot, the folded desk lamp unfolds through its real hinges: the base stays fixed, the stem rises, the arm extends, and the shade rotates into its final position. The lamp switches on only during the final second. Preserve the lamp's color, materials, proportions, desk surface, framing, and background throughout. Smooth physically plausible mechanical motion, no cuts, no camera movement, no new parts, no disappearing parts, no text.Settings: First and last frames · 6 seconds · 720p · recreated endpoint crops · audio removed for publication
What the result showed: The lamp unfolded smoothly through a coherent mechanical sequence while the base, tabletop, and camera remained stable. The light also became visibly brighter near the end, making this the clearest transformation example in the set.
Example 4: Motion Transfer From a Video Reference
Example 4 reuses the unedited greenhouse-tram output from Example 1. The prompt asks Seedance 2.5 to retain its camera direction, subject speed, and pacing while replacing the scene and subject.
Reference video:
t2v eg1.mp4
Recreation prompt
Use @video1 only as the reference for camera movement, subject speed, and pacing. Replace the yellow tram and greenhouse with a compact white autonomous delivery robot rolling left to right through a clean covered night-market corridor after rain. Do not reproduce the tram, glass greenhouse, tropical plants, or yellow-and-blue color palette from @video1. Track parallel to the robot at the same approximate height and distance as the reference. Keep all six wheels rotating naturally and the robot's white body geometry stable. Reflections move across the wet floor as the camera travels. One continuous shot, realistic night lighting, no cuts, no text, no people.Settings: Omni reference · 5 seconds · 720p · 16:9 · audio off
What the result showed: The new clip retained a similar lateral tracking rhythm while replacing the tram and greenhouse with a white robot and wet market corridor. The scene did not copy the original setting, and the moving floor reflections reinforced the camera movement.
Example 5: Reference-Guided Desk Lamp Relighting
The final example uses the supplied desk-lamp product sheet as an identity reference, then changes the setting and lighting while preserving the product design.
Recreation prompt
Use @image1 as the exact lamp product reference. Preserve the lamp's identity, silhouette, materials, ivory-and-sage colors, proportions, circular base, and hinge details. Present the open lamp on a gray stone pedestal against softly illuminated pale-blue translucent panels with cool diffused daylight. Keep the product stable while the camera moves smoothly toward a centered commercial composition. No mist, extra products, hands, added text, cuts, or audio.Settings: Reference-guided generation · 5 seconds · 720p · actual 16:9 · audio off
What the result showed: The result is a clean lamp product shot with cool lighting, pale-blue panels, and stable motion. The ivory body, sage accents, circular base, and articulated shape remain recognizable from the supplied product reference.
What the Five Seedance 2.5 Prompts Taught Us
The examples point to five practical prompt-writing habits:
- Describe one shot, not a sequence of edits. All five prompts use continuous motion that fits within five or six seconds.
- Separate subject motion from camera motion. “The tram glides” and “the camera tracks parallel” give the model two distinct instructions.
- Give each reference one role. Example 2 uses an image for product identity, while Example 4 uses a video for movement and pacing.
- Name what must stay fixed. The lamp example explicitly holds the base, camera, tabletop, and background stable during the transformation.
- Check effective output shape. A reference image can override the request aspect ratio, so prepare the source asset in the format you intend to publish.
Limitations observed: Foliage obscured a requested mechanical detail in Example 1, Example 2 returned a horizontal composition instead of the planned vertical ad, and Example 4 showed four visible wheels instead of the requested six. Inspect every delivered file before adapting a prompt for production.
How to Recreate a Seedance 2.5 API Example
Send an authenticated task to POST https://api.piapi.ai/api/v1/task with model seedance and task type seedance-2.5. Keep your API key on the server and use publicly accessible URLs for any supplied references.
curl --request POST "https://api.piapi.ai/api/v1/task" \
--header "X-API-Key: $PIAPI_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"model": "seedance",
"task_type": "seedance-2.5",
"input": {
"prompt": "A small yellow maintenance tram glides through a rain-soaked glass greenhouse at blue hour. The camera tracks parallel in one continuous shot. Warm interior light contrasts with the cool greenhouse. No cuts, text, or people.",
"mode": "text_to_video",
"duration": 5,
"resolution": "720p",
"aspect_ratio": "16:9",
"audio": false
}
}'For reference-guided prompts, add image_urls, video_urls, or audio_urls and mention each asset in the prompt as @image1, @video1, or @audio1. The API currently supports up to 9 images, 3 videos, and 3 audio files. See the Seedance 2.5 request contract for validation rules and response details.
Seedance 2.5 Prompt FAQ
Can I copy these Seedance 2.5 prompt examples?
Yes. Copy the structure, then replace the subject, action, environment, and reference roles with details from your own project. If a prompt uses @image1 or @video1, supply the corresponding media URL in the request. Results can vary between generations, so review the returned video before using it in production.
Can Seedance 2.5 use image, video, and audio references?
Yes. PiAPI's Seedance 2.5 task supports image, video, and audio references through image_urls, video_urls, and audio_urls. The current limits are 9 images, 3 videos, and 3 audio files. An audio reference requires at least one image or video reference in the same request.
How long can a Seedance 2.5 video be on PiAPI?
The current PiAPI documentation accepts integer durations from 4 to 15 seconds, with 5 seconds as the default. Because product pages and older articles can retain earlier limits, use the live API documentation as the source of truth when building or updating an integration.
Does Seedance 2.5 support 1080p?
No. PiAPI currently supports 480p and 720p for the seedance-2.5 task type, with 720p as the default. Requests for 1080p are not supported under the documented contract. Plan your generation and publishing workflow around 480p or 720p outputs.
How much did these five Seedance 2.5 examples cost?
The five examples contain 26 requested output seconds at 720p. At the documented PiAPI pricing rate of $0.60 per output second, the output-only total is $15.60. Example 4 also uses a 5.08-second video reference; applying the documented half-rate input-video formula adds about $1.52, for an estimated total of about $17.12.
Try a Seedance 2.5 Prompt on PiAPI
These Seedance 2.5 prompts show how text, images, endpoint frames, and motion references can guide different kinds of short video. Start with one clear subject and one achievable shot, then add only the reference and preservation instructions your workflow needs.
Try one of these prompts in the Seedance 2.5 playground, or review the Seedance 2.5 API documentation before integrating the task into your application.
- Five audio-free publication MP4s and poster images are implemented.
- The supplied watermelon and desk-lamp inputs are implemented as WebP assets.
- Example 3 uses transparent recreation crops prepared from the supplied folded/open lamp product sheet.
- The article-specific 2000×1000 WebP cover is implemented.
- The prompts are labeled as recreation prompts rather than task-log transcripts.
*Sources checked August 3, 2026: Seedance 2.5 Preview API documentation, Seedance 2.5 API specifications, and Seedance 2.5 model and playground guide. The live API documentation is the source of truth for the current request contract.*
