What image formats and size does image_url accept?

image_url takes a public, directly downloadable link to a JPEG, PNG or WebP file under 10 MB. It is required alongside prompt and it is the only image field on this endpoint, so whatever sits at that URL becomes the first frame exactly as it is. Text, watermarks and logos in the picture travel into the video, so clean the source first.

Can Seedance 1.5 Pro image-to-video make a portrait speak?

Yes. Send the portrait as image_url, leave generate_audio on, and write the spoken line and its delivery into the prompt, since the soundtrack is rendered with the picture in one pass. ByteDance describes the model as generating voices with lip-sync across a range of languages and dialects. Keep the line short enough to fit inside four to twelve seconds.

What happens if my image's aspect ratio does not match aspect_ratio?

The clip is fitted to the value in the request. Seedance 1.5 Pro publishes no adaptive framing option, so aspect_ratio decides the output shape and a source with a different crop is reframed to match it, which can take margins off the subject. Match the still's crop to the ratio you intend to deliver, or expect the edges to go.

How do I keep the same character across several clips?

Reuse the same image_url to carry the character’s face, wardrobe and lighting into each clip. Keep seed fixed when comparing motion prompts, and save both fields with each selected take.

Is a text prompt required as well as the image?

Yes. Seedance 1.5 Pro image-to-video requires prompt and image_url together, with the prompt at least three characters long, because the still supplies the scene and the prompt supplies the motion. A request carrying only an image fails validation. Short and specific beats a re-description of the picture: name the movement, the camera behaviour and the ending.

Why does a longer clip drift away from my source image?

Length is the usual cause. Seedance 1.5 Pro animates forward from one frame with nothing pinning the end, so the further a clip runs from that frame, the more the picture can travel away from it. Confirm the motion at four or five seconds, extend only once it holds, and keep camera_fixed on when the framing itself has to stay put.

Can the opening frame be a generated image rather than a photograph?

Yes, as long as it sits at a public URL in an accepted format. image_url does not care where the picture came from, so a frame from an image model can go straight in as the opening frame, which is a common way to settle the look first and put it in motion second. The 10 MB ceiling applies either way.

Does one request return more than one clip?

No. Seedance 1.5 Pro image-to-video returns one clip per request, because the endpoint has no batch or count field. Run variants as separate requests, each with its own duration_seconds, which is the quantity the request is billed on. The task id you get back identifies that single render, so polling it never yields a second version.

Seedance 1.5 Pro API — Uncensored NSFW Video, Locked Camera

bytedance/seedance-1.5-pro/image-to-video

Seedance 1.5 Pro image-to-video from ByteDance starts the clip on a still you supply and animates it. Clips run 4 to 12 seconds at 480p, 720p or 1080p with native audio, and a camera lock keeps the framing exactly as composed.

From $0.011/s

Run model View API

Input

Image to video

Prompt

Input image1/1

Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF

Closing frame0/1

Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF

Aspect ratio

Resolutionaffects price

Durationaffects price

Generate audioaffects price

Lock camera

Seed

Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run

Example result

Extend

Video editing

Video upscaling

{
  "model": "bytedance/seedance-1.5-pro/image-to-video",
  "input": {
    "aspect_ratio": "3:4",
    "resolution": "720p",
    "duration_seconds": 6,
    "generate_audio": false,
    "camera_fixed": false,
    "prompt": "Locked-off camera: bare-chested, he works the heavy bag, sweat flying off on the impacts, the bag swinging back into frame each time while the background stays dead still. Sweat sheets off the bare back and shoulders on every impact; close, hard side light. Cinematic, hard side light, heavy film grain, gym green and black.",
    "image_url": "https://cdn.spicyapi.ai/models/examples/inputs/a89f0ee3314c7a0b.webp"
  }
}

Introduction

Seedance 1.5 Pro image-to-video starts the clip on a picture you supply and animates it for up to 12 seconds. With camera_fixed set, the supplied framing stays exactly as composed, which is why this endpoint produces usable plates.

ByteDance Seedance 1.5 Pro publishes two endpoints, text-to-video and image-to-video. Both return 4 to 12 second clips at 480p, 720p or 1080p with native audio, and both carry a camera_fixed switch.

Choose a Seedance 1.5 Pro mode by starting material

Starting material	Mode
An approved opening still	Seedance 1.5 Pro Image to Video
A script or scene description	Seedance 1.5 Pro Text to Video

Key features & highlights

camera_fixed holds the frame

Left false, the default, the model invents its own moves: a drift, a push, a slow orbit. Set it to true and the camera stops while the subject keeps moving inside the frame. A locked frame can be tracked, keyed, composited or looped, and neither Seedance 2.0 nor 2.5 offers this switch.

Sound in the same pass

generate_audio defaults to true and renders ambience, movement effects and voice locked to the frames. It moves the request into a higher rate tier, so turn it off for product loops and for plates that will be scored later.

A clear ceiling

Clips run 4 to 12 whole seconds at 480p, 720p or 1080p, in six fixed aspect ratios with 16:9 as the default and no adaptive option. There is no reference endpoint; seed is available at 480p and 720p. Jobs that need 4K, longer takes or a multi-image cast belong on Seedance 2.0 or 2.5; what 1.5 Pro keeps is the camera lock.

Request parameters

Request parameters for Seedance 1.5 Pro · Image to Video
Parameter	Required	Description
`prompt`	Yes	Required, and it should say what happens after the frame rather than restate what is already visible: the action, the passage of time, the sound. at least 3 characters
`image_url`	Yes	Public, directly downloadable link to the first frame; required.
`last_image_url`	No	Optional closing frame: send it with `image_url` and the shot has to land on this still, which makes transitions and before-and-after pairs repeatable.
`aspect_ratio`	No	Output framing, six fixed values, 16:9 by default. `21:9` · `16:9` · `4:3` · `1:1` · `3:4` · `9:16` · Default: `16:9`
`resolution`	No	Output resolution, 720p by default and 1080p at the top; nothing higher is offered here. It sets a rate tier. `480p` · `720p` · `1080p` · Default: `720p`
`duration_seconds`	No	Length in whole seconds, 4 to 12, defaulting to 5, and the billed quantity. `≥ 4` · `≤ 12` · Default: `5`
`generate_audio`	No	Whether a native soundtrack is rendered in sync with the picture. On by default and priced in a higher tier; switch it off for silent product loops or when sound is added in post. Default: `true`
`camera_fixed`	No	Locks the camera, false by default. Set to true and the frame holds still while the subject moves within it — ideal for product shots and plates destined for a static background. Default: `false`
`seed`	No	Integer from 0 to 2147483647 for the random start; leave it out and every run draws its own. `≥ 0` · `≤ 2147483647`

Pricing

Pricing
Resolution	Generate audio: No	Generate audio: Yes
`480p`	$0.0112 / second	$0.0223 / second
`720p`	$0.0227 / second	$0.0454 / second
`1080p`	$0.0491 / second	$0.0982 / second

Use cases

Plates for compositing

Send a still you have already approved, set camera_fixed, and the composition holds for the whole clip while only the subject moves. The result behaves like tripod footage: a clean background to key against, a stable surface to track a title onto, a plate an editor can cut into a real sequence.

Product turntables and loops

A fixed frame with a rotating object is the classic case: nothing drifts, the horizon stays level, and the last frame lands close enough to the first to loop. Turn generate_audio off, because a product loop almost never wants room tone.

Slow environmental motion

The published example layers falling snow, rolling steam and water over a single still, with a very slow push-in so none of it is hidden by camera speed. Weather, smoke and water are the kind of motion a held composition shows off best.

Explore key use-case tutorials

Prompt guide & pro tips

Send both required fields

prompt and image_url are both required here, and the prompt needs at least 3 characters. The image must be a public, directly downloadable JPEG, PNG or WebP under 10 MB. Text, watermarks and logos in the still travel into the video, so clean them off first.

Match the aspect ratio to the still

There is no adaptive option, and aspect_ratio defaults to 16:9. When the requested ratio differs from the source, the frame is reframed and the subject can lose its margins, so pick the ratio that matches your still.

Adapt the included example

The published example uses aspect_ratio 16:9, duration_seconds 5 and resolution 720p, with the audio default left on.

Snow keeps falling through the lantern light and steam pours off the water. She sets the folded towel robe down on the rock and sinks slowly into the spring up to her collarbone, tipping her head back and exhaling a long plume of breath into the cold air. Steam swells across the frame and snow settles on her wet hair. Very slow push-in, warm lantern gold against deep blue snow.

The frame supplies the lanterns and the setting; the prompt only lowers her into the spring and lets her breathe out. Keep that split when you swap in your own still.

Keep motion inside a locked frame

With camera_fixed set to true, describe only movement that happens within the frame, because a camera instruction and a locked camera pull against each other. Confirm the motion at 4 or 5 seconds before extending to 12, since a longer shot drifts further from its source frame.

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-1.5-pro/image-to-video",
  "input": {
    "aspect_ratio": "16:9",
    "duration_seconds": 5,
    "generate_audio": true,
    "image_url": "https://example.com/reference.png",
    "prompt": "A slow aerial shot drifting over a misty pine forest at sunrise",
    "resolution": "720p"
  }
}'

# 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.

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Animate one frame you already approved

Seedance 1.5 Pro's image-to-video endpoint takes image_url as the opening frame and animates it for 4 to 12 seconds at up to 1080p, with sound rendered in the same pass. The still fixes composition, likeness and grade, so the prompt only has to say what moves. Use last_image_url with image_url at 480p or 720p when the shot needs a defined closing frame.

Open the playground

Required input: prompt and image_url (the opening frame)
Closing frame: last_image_url · 480p / 720p
Duration: 4-12 whole seconds (duration_seconds), default 5
Resolution: 480p / 720p / 1080p, default 720p
Audio: generate_audio renders sound in the same pass, true by default
Reruns: Optional seed: integer from 0 to 2147483647 for controlled comparisons

One anchor: image_url starts it, nothing pins the endThis endpoint accepts a single picture. The clip begins on that frame and the ending is described in the prompt, so a shot that has to land on one exact final image is better planned as two takes cut together than as a single request.
Identity: The opening frame owns who is in itWardrobe, framing, grade and likeness all arrive with the picture, which also decides how good the clip can look, since nothing downstream sharpens a soft source. The prompt handles what happens next: the movement, the camera behaviour, and where the shot finishes.
Ceiling: The same limits as text-to-videoThe boundaries match the text endpoint: one image in, four to twelve seconds out, 1080p at the top and no reference-to-video companion on this generation. Use last_image_url with image_url at 480p or 720p when the shot needs a defined closing frame.

What the Seedance 1.5 Pro API does well

Camera lock still applies

camera_fixed behaves here exactly as it does on text-to-video: set it and the framing you approved in the still is the framing the clip keeps, while the subject moves inside it. That combination is what turns a photograph into a plate for tracking or compositing, so watch the first take for drift before an edit is built on top of it.

Synced audio by default

generate_audio defaults to true, so an animated still comes back with sound attached. ByteDance describes the model as producing voices and spatial effects that coordinate with the visuals, with lip-sync across a range of languages, which is what makes a portrait plus a written line a usable talking-head take. Switch it off for silent loops, and remember audio selects a higher rate tier.

Six ratios, matched to the source or reframed
The same six values as text-to-video apply: 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, defaulting to 16:9. Nothing reads the picture and adapts to it, so a still whose crop differs from aspect_ratio is fitted to the value you asked for and the subject can lose its margins. Crop the source first when the framing matters.
A fair way to compare motion variants
Keep the prompt, seed, duration and resolution unchanged, then vary one setting at a time. On image-based tasks, reuse the same source images. Compare the staging and motion across a few takes to see whether the change helps.
A smaller contract, easier to reason about
Pair prompt with image_url for the opening shot, add last_image_url at 480p or 720p when the ending matters, and use seed for controlled comparisons. Set duration_seconds, resolution, aspect_ratio, generate_audio and camera_fixed to suit the shot.
Check at 480p, deliver at 1080p
Refine the shot at the lower resolution, then render the selected version at the delivery resolution with the same inputs and seed. Changing resolution starts a new generation, so review the final staging and motion before delivery.

Who Seedance 1.5 Pro is for

Product and interview shots
Approve the still, then set only the motion
Start from a photograph that has already cleared review, turn camera_fixed on, and write a prompt that names the one thing that should move. None of the composition is up for renegotiation, which is what makes this route predictable for listings, interviews and anything that has to match a still already in use.
Short-form social cuts
Match the still to the destination ratio first
Crop or shoot the source picture in the ratio the feed expects before generating, because a mismatch is resolved by fitting the image to the requested value rather than by adapting to what you sent. Vertical placements usually want the subject centred with headroom to spare, since a reframe takes the difference off the edges.
First-pass exploration under 12 seconds
Animate one clear beat from a still
Four to twelve seconds is the range, so describe a single action: a turn, a gesture, a breath, a step into frame. Confirm it at four or five seconds first, because the longer a clip runs, the further the picture travels from the frame you started with.
Localisation and dialogue tests
One approved still, several spoken lines
Keep image_url fixed and vary only the line in the prompt when the same presenter has to read different copy. ByteDance describes the model as supporting several languages and dialects, so a test render per language is how you find out what carries before a batch is committed.

How to call the Seedance 1.5 Pro API

Approve the frame before animating it
Whatever is wrong in the still is wrong in every frame of the clip, so fix exposure, crop and stray text first. Host the file at a public, directly downloadable URL as JPEG, PNG or WebP under 10 MB, since image_url takes a link rather than file bytes.
Decide the camera before writing the rest of the prompt
camera_fixed changes what the prompt has to carry. A held frame needs only the subject's movement described, while a free camera needs a direction and a pace for the move as well as the action happening inside the frame.
Write what happens after frame one
The picture already states what is there. Spend the prompt on what moves, how the camera behaves and how the shot should look at the end, since no field pins the ending down and the wording is the only thing that decides it.
Set audio deliberately
generate_audio is on unless you turn it off, and it selects a higher rate tier. Leave it on when a spoken line or ambience ships with the clip, and switch it off when the still is becoming a silent loop or a plate for a sound pass.

Seedance 1.5 Pro worked examples

A rotating product shot from a single photo
image_url is the approved hero shot, camera_fixed true, 8 seconds — the product turns in place while the frame itself stays put.
```
Slowly rotate the product 180 degrees in place, soft studio lighting stays constant, no camera movement, gentle turntable motion only
```

A talking-head beat animated from a still portrait

image_url is an approved portrait, camera_fixed true, audio on, and the prompt carries the ending because no field can set it.

The person in the portrait speaks calmly to camera, subtle breathing and blinking only, camera does not move, settles back to a still, closed-mouth expression by the last frame

A wide outdoor still brought to life with camera motion
image_url is a still photograph, camera_fixed false, 10 seconds, 21:9, so the prompt directs a camera move rather than holding the frame the way the two shots above do.
```
Slow push forward over the still lake in the photo, ripples spread outward as a light breeze picks up, mist drifts across the far shore, camera continues moving forward through the whole shot
```

Seedance 1.5 Pro API questions

Treat image_url as the fact and the prompt as the instruction: say what happens after the frame, not what the frame already shows. The ending has to be written in words, since no field pins it down, and a short clip keeps that ending within reach. When the approved framing has to survive the whole shot, turn camera_fixed on before worrying about the wording.

Workbench