Introduction
Seedance 2.0 Fast has no separate edit switch: it reads your clip as its first reference video and works out from the wording that the footage should be reworked rather than imitated. Name the change, say what stays, and the result returns at the source's length and frame shape.
| Field | Value |
|---|---|
| Input | Video |
| Output | Video |
| Required input | prompt, |
| Source clip | 2–15 s · MP4 · 409,600–2,086,876 px per frame · aspect 0.4–2.5 |
| Output length | Matches the source; clips under 4 s return about 4 s |
| Resolution | 480p, 720p (default); frame shape follows the source |
| References | Up to 9 images and 3 audio tracks, optional |
| Audio |
|
| Billing | Per source second, rounded up, 4 s minimum |
Key features & highlights
Choose the Fast mode for your source material
A sentence goes to text-to-video. An approved first frame goes to image-to-video, which follows that frame far more literally than the same words typed as a prompt. A face, a garment or a room you need held steady goes to reference-to-video. Choosing by what you already own is what keeps a request from silently ignoring half its fields.
Let the source set the shape
aspect_ratio defaults to adaptive: image-to-video keeps the proportions of the opening frame and reference-to-video follows the primary reference, so a portrait still is not cropped into a landscape box. The six fixed ratios — 16:9, 9:16, 4:3, 3:4, 1:1 and 21:9 — are still there when a placement dictates the shape.
Nine references, three clips, three tracks
reference-to-video reconciles up to 9 images, 3 video clips and 3 audio tracks in a single request. Three angles of one face hold a character far better than nine unrelated stills, so give each asset one job instead of asking all of them to carry identity, style and motion at once.
Request parameters
| Parameter | Required | Description |
|---|---|---|
prompt | Yes | Lead with the change, then list what must not move: faces, hands, props, the timing of each gesture and the camera path. at least 1 characters |
video_url | Yes | MP4, 2 to 15 seconds, 24 to 60 fps, 50 MB at most. at least 1 characters |
reference_image_urls | No | Optional: up to 9 stills that pin down the target look, such as a jacket, a prop, a face or a colour board. at least 1 items · at most 9 items |
reference_audio_urls | No | Optional: up to 3 files of 2–15 seconds, 15 seconds combined, that steer the new soundtrack toward a voice or a mood. at least 1 items · at most 3 items |
resolution | No | 480p or 720p (default). The frame shape follows the source: a widescreen clip came back 864×496 at 480p in our tests, a square one 640×640. |
generate_audio | No | On by default: Fast writes a new soundtrack for the edited picture. Default: |
seed | No | Optional integer from 0 to 2147483647. Holding it fixed makes two versions of one edit easier to compare, though it does not promise matching frames. |
return_last_frame | No | Also returns the edited clip's closing frame as a separate image, handy as the opening still of a follow-up shot. Off by default. Default: |
bitrate_mode | No | standard (default) or high. High encodes the delivered file at a higher bitrate for crisper detail and a larger download; the price stays the same. |
watermark | No | Off by default. Turn it on to burn the model's visible "AI generated" mark into the edited clip. Default: |
Pricing
Charged per second of the source clip, rounded up, with a 4-second minimum; the chosen resolution sets the rate. Reference images, reference audio and the audio switch leave the price unchanged, and a task that fails costs nothing.
| Specification | Price |
|---|---|
| $0.045078 / second |
| $0.096942 / second |
Use cases
Vertical cuts for a social feed
Seedance 2.0 Fast text-to-video turns a written shot into 4–15 seconds of footage; specify the camera movement and sound before spending another render on a revised idea.
Uncensored shorts, couples included
SpicyAPI adds no review to Fast requests, but earlier SpicyAPI tests recorded model-side refusals of some text prompts; the unrestricted label is not a promise that every adult NSFW request succeeds.
Extend a shot from its last frame
SpicyAPI’s Seedance 2.0 Fast video-extend uses the source clip’s last frame to generate a continuation and joins the two segments for delivery.
Draft at 480p, deliver at 720p
Fast exposes 480p and 720p, with 720p selected initially; use the smaller frame for motion drafts and check fine detail again at the delivery size. Fast exposes an optional integer seed from 0 to 2147483647; keeping it fixed helps compare one edit at a time but cannot guarantee identical frames.
Prompt guide & pro tips
Trim the Seedance 2.0 Fast source
Cut the footage down to the 2–15 seconds that need changing, check it sits between 480p and 1080p, and send it as
video_url.Word the edit for Fast
Begin with "Edit @Video1:", name one change, then list what has to hold: faces, gestures and their timing, props and the camera move.
Review the Fast edit against the source
Play both side by side and step through hands, faces and fine textures; once the 480p version reads right, rerun the same prompt at 720p.
What Seedance 2.0 Fast video edit does
Seedance 2.0 Fast is the quicker, lower-cost tier of ByteDance's Seedance 2.0 video model, built for turning takes around fast. This endpoint points it at footage you already have: the source clip goes in as video_url, the model reads it as @Video1, and your prompt says what should look different. Who is in the shot, how they move and where the camera travels come from the source; wardrobe, light, weather, set dressing or the overall look come from your words.
Seedance 2.0 has no dedicated edit switch. It decides from the instruction whether a reference clip should be followed loosely or reworked frame for frame, so the request locks length and frame shape to the source and leaves the rest to the prompt. Write it as an edit ("Edit @Video1: …") and spell out what has to stay.
Request example
This is the izakaya edit from our tests. It goes to POST https://api.spicyapi.ai/api/v1/jobs/createTask with an Authorization: Bearer YOUR_SPICY_API_KEY header; the task API guide covers polling and callbacks.
{
"model": "bytedance/seedance-2.0-fast/video-edit",
"input": {
"prompt": "Edit @Video1: change her white button-down shirt into a black silk camisole with thin straps, and turn the warm kitchen light into a deep red neon glow that reflects on the steel grill and in the rising smoke. Keep both faces, their long look across the grill, the steam, every movement and its timing, and the camera exactly as in @Video1.",
"video_url": "https://example.com/izakaya.mp4",
"resolution": "480p",
"generate_audio": true
}
}
The call returns a task ID straight away. Poll it, or pass callBackUrl, and the finished task lists the edited MP4, plus its closing frame when return_last_frame is on.
Source clip rules
- Length from 2 to 15 seconds; anything outside that window is refused before a charge is made.
- MP4 at 24–60 fps, no larger than 50 MB.
- Between 409,600 and 2,086,876 pixels per frame, with an aspect ratio from 0.4 to 2.5. In practice 480p, 720p and 1080p footage works, while 360p clips and anything sharper than 1080p are turned away without a charge.
- Faces and hands hold up best when they fill a reasonable part of the frame; the edit cannot invent detail the source never had.
Writing the edit instruction
Open with the change and make it concrete ("her white shirt becomes a black silk top", "the afternoon turns into a red-lit night"). Then list what must hold: faces, hair, every gesture and its timing, props, the set and the camera path. One or two related changes per run land more reliably than a long list; run a second edit on the result when you need more.
Reference stills help when words cannot pin a look down. Attach up to nine as reference_image_urls and point at them ("swap his jacket for the one in @Image1"). Reference audio, up to three files and 15 seconds in total, steers the new soundtrack and only matters while generate_audio is on.
Sound, resolution and length
generate_audio starts on and writes a new soundtrack for the edited picture; in the example, grill hiss, bar chatter and a bass line replaced the original sound. Switched off, the edit is silent, since the source's own audio is not carried across.
resolution offers 480p or 720p, and the frame shape follows the source: our 16:9 sources came back 864×496 at 480p, a square one 640×640. Length follows the source too, plus one frame (4.00 s in, 4.04 s out; 5.00 s in, 5.09 s out). Fast never renders less than 4 seconds, so a 2-second clip is edited for its 2 seconds and then continued to about 4.
How billing works
The charge is per second of the source clip, rounded up to whole seconds with a 4-second minimum, at the rate for the chosen resolution. A 5-second clip bills as 5 seconds, a 4.2-second clip as 5, and a 2-second clip as 4. Reference images, reference audio and the audio switch leave the price unchanged. For clips uploaded to SpicyAPI the estimate shown before you submit is exactly what you pay; for external links the hold covers 15 seconds and the rest is released when the edit finishes. Failed or rejected tasks are not charged.
What our tests showed
Before launch we ran five 480p edits and compared each with its source frame by frame. All five changed only what the prompt named (a wardrobe swap, a relit set, a new background) while the people, their movements and the camera lined up with the original. Edits at the Moderate and Strong content levels rendered as written, and none was refused. The 2-second source returned 4.04 seconds long: its first 2 seconds followed the source and the rest carried the shot on.
Edit only footage you shot or hold the rights to change, featuring adults who agreed to appear in it; Seedance 2.0 Fast must not be used to drop a real person into a misleading or non-consensual scene.
Edit only footage you shot or hold the rights to change, featuring adults who agreed to appear in it; Seedance 2.0 Fast must not be used to drop a real person into a misleading or non-consensual scene.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "bytedance/seedance-2.0-fast/video-edit",
"input": {
"bitrate_mode": "standard",
"prompt": "A slow aerial shot drifting over a misty pine forest at sunrise",
"resolution": "720p",
"video_url": "https://example.com/reference.mp4"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.










