Seedance 2.5 Omni Reference (Intl): Image-to-Video

Blend images, video, and audio into one video with Seedance 2.5 Omni Reference, resolution 480p–4K. International-region (Intl) endpoint. Seedance 2.5 Omni Reference blends multiple reference images, video clips, and audio into a single guided generation, with selectable output resolution (480p/720p/1080p/4K) — price scales with the resolution chosen. Supports clips up to 30 seconds. This international-region endpoint is served via a Dreamina-hosted deployment of the same Seedance 2.5 model, for traffic outside mainland China.

Interactive model controls

Seedance 2.5 Omni Reference blends multiple reference images, video clips, and audio into a single guided generation, with selectable output resolution (480p/720p/1080p/4K) — price scales with the resolution chosen. Supports clips up to 30 seconds. This international-region endpoint is served via a Dreamina-hosted deployment of the same Seedance 2.5 model, for traffic outside mainland China.

📝

Overview

About this model

Seedance 2.5 Omni Reference is the multi-reference endpoint in the Seedance 2.5 family, blending up to 30 reference images, 10 video clips, and 10 audio files into a single guided video generation, with selectable output resolution (480p/720p/1080p/4K) — price scales with the resolution chosen. Use the images to steer environment and style, the videos for camera motion and rhythm, and the audio as a mood reference — all combined via one text prompt. Supports clips up to 30 seconds. This is Muapi's international-region endpoint for the same Seedance 2.5 model, served via Seegen's Dreamina deployment for traffic outside mainland China. This endpoint also supports Draft Mode: set draft: true for a fast, low-cost 480p preview (resolution is ignored — sending anything other than 480p alongside draft=true is rejected), then call this same endpoint again with draft_request_id set to that draft's request_id to finalize it at full 1080p, reusing the prompt and reference materials from the original draft automatically. A draft stays finalizable for 7 days after it completes.

1Style-consistent content: Combine multiple reference images to lock in a consistent visual style across shots.

2Music-driven video: Use an audio clip as a mood and pacing reference for the generated motion.

3Camera motion transfer: Reference an existing video clip to carry over its camera movement into a new scene.

4Multi-source composition: Merge environment images, motion references, and audio mood into one coherent clip.

💰

Pricing & Value

Cost analysis

Provider	Cost	Notes
muapiapp	$0.187/sec (480p), $0.374/sec (720p), $0.935/sec (1080p), or $1.87/sec (4K)	Rate applies to output duration only when no video reference is used, selectable via the `resolution` parameter on this same endpoint. With a video reference, billed at 65% of that rate on the combined duration of the reference videos plus the generated output.
Fal.ai	$0.47/sec (720p) or $0.2205/sec (480p)	MuAPI is cheaper than Fal.ai at 480p; comparable at 720p once the Intl surcharge is factored in.
Replicate	Not available	Seedance 2.5 is not listed on Replicate.

muapiapp$0.187/sec (480p), $0.374/sec (720p), $0.935/sec (1080p), or $1.87/sec (4K)

Rate applies to output duration only when no video reference is used, selectable via the `resolution` parameter on this same endpoint. With a video reference, billed at 65% of that rate on the combined duration of the reference videos plus the generated output.

Fal.ai$0.47/sec (720p) or $0.2205/sec (480p)

MuAPI is cheaper than Fal.ai at 480p; comparable at 720p once the Intl surcharge is factored in.

ReplicateNot available

Seedance 2.5 is not listed on Replicate.

** Competitor pricing is estimated based on similar model architectures and usage tiers.

⚙️

Technical Details

Configuration schema

Parameter	Type	Description	Default
Prompt	string	Text prompt describing the desired video, referencing the provided images, video clips, and audio as environment, motion, and mood cues.	`Create a calm cinematic park sequence. Use the images for environment style, the video clips for camera motion and street rhythm, and the audio as mood reference.`
Reference Images	array	Reference image URLs. Up to 30 images.	`undefined`
Reference Videos	array	Reference video URLs. Up to 10 clips.	`undefined`
Reference Audio	array	Reference audio URLs. Up to 10 files.	`undefined`
Resolution	Enum (4 options)	Output resolution. Price scales with resolution: 480p is cheapest, 4K is most expensive.	`720p`
Duration	int	The duration of the generated video in seconds.	`5`
Aspect Ratio	Enum (8 options)	Aspect ratio of the output video.	`16:9`
Omni Reference Task Type	Enum (4 options)	Hint for the omni-reference subtask type, so ratio/duration constraint mismatches are caught at submission time instead of failing asynchronously. auto lets the model infer the type from the prompt; reference has no special ratio/duration constraints; edit and extend both require ratio=adaptive (edit additionally requires duration=-1). The model still re-derives the actual task type from the prompt during processing, so a mismatch can still surface as an async error.	`auto`
Seed	int	Random seed for reproducible generation. Use -1 for random.	`42`
High Bitrate	boolean	Enable high bitrate mode for better visual fidelity. Produces larger files.	`false`
Draft	boolean	Generate a cheap 480p draft (resolution must be 480p or omitted when true — any other value is rejected). Finalize the draft you keep at 1080p by calling this same endpoint again with draft_request_id set to this request's request_id.	`false`
Draft Request ID	string	Finalize an earlier draft (submitted with draft=true) into a full 1080p render. Set to that draft's request_id; prompt, media, duration, aspect ratio and seed are reused from the draft. resolution must be 1080p or omitted, and draft must not be true — anything else is rejected. The draft must be completed and at most 7 days old.	`-`

Promptstring

Text prompt describing the desired video, referencing the provided images, video clips, and audio as environment, motion, and mood cues.

Default Value

Create a calm cinematic park sequence. Use the images for environment style, the video clips for camera motion and street rhythm, and the audio as mood reference.

Reference Imagesarray

Reference image URLs. Up to 30 images.

Default Valueundefined

Reference Videosarray

Reference video URLs. Up to 10 clips.

Default Valueundefined

Reference Audioarray

Reference audio URLs. Up to 10 files.

Default Valueundefined

ResolutionEnum (4 options)

Output resolution. Price scales with resolution: 480p is cheapest, 4K is most expensive.

Default Value720p

Durationint

The duration of the generated video in seconds.

Default Value5

Aspect RatioEnum (8 options)

Aspect ratio of the output video.

Default Value16:9

Omni Reference Task TypeEnum (4 options)

Hint for the omni-reference subtask type, so ratio/duration constraint mismatches are caught at submission time instead of failing asynchronously. auto lets the model infer the type from the prompt; reference has no special ratio/duration constraints; edit and extend both require ratio=adaptive (edit additionally requires duration=-1). The model still re-derives the actual task type from the prompt during processing, so a mismatch can still surface as an async error.

Default Valueauto

Seedint

Random seed for reproducible generation. Use -1 for random.

Default Value42

High Bitrateboolean

Enable high bitrate mode for better visual fidelity. Produces larger files.

Default Valuefalse

Draftboolean

Generate a cheap 480p draft (resolution must be 480p or omitted when true — any other value is rejected). Finalize the draft you keep at 1080p by calling this same endpoint again with draft_request_id set to this request's request_id.

Default Valuefalse

Draft Request IDstring

Finalize an earlier draft (submitted with draft=true) into a full 1080p render. Set to that draft's request_id; prompt, media, duration, aspect ratio and seed are reused from the draft. resolution must be 1080p or omitted, and draft must not be true — anything else is rejected. The draft must be completed and at most 7 days old.

Default Value-

📖

Implementation Guide

Developer documentation

How to Use Seedance 2.5 Omni Reference

Gather Your References
- Up to 30 images via images_list, 10 videos via videos_list, 10 audio files via audios_list. At least one reference is recommended.
Write a Guiding Prompt
- Describe how the references should combine.
- Example: "Create a calm cinematic park sequence. Use the images for environment style, the video clips for camera motion and street rhythm, and the audio as mood reference."
Configure Output
- Set resolution to 480p, 720p (default), 1080p, or 4k — price scales with resolution (see the pricing table below).
- Set duration (4–30s), aspect_ratio, and optionally seed.
Submit via the API

curl -X POST https://api.muapi.ai/api/v1/seedance-2.5-intl-omni-reference \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Create a calm cinematic park sequence using the references for style, motion, and mood.",
    "images_list": ["https://example.com/ref1.jpg", "https://example.com/ref2.jpg"],
    "videos_list": ["https://example.com/ref.mp4"],
    "resolution": "720p",
    "duration": 10
  }'

Poll and Download
- Poll GET /api/v1/predictions/{request_id}/result until status is completed, then retrieve the video URL from output.

❓

Common Questions

Frequently asked

What's different about the Intl endpoint?

Nothing about the request/response shape — it's the same Seedance 2.5 model, served through Seegen's Dreamina deployment for international-region traffic instead of the mainland-China Doubao deployment. Pricing carries a 10% surcharge over the equivalent Doubao-tier rate.

What does Seedance 2.5 Omni Reference do?

It generates video guided by a combination of reference images, video clips, and audio files, blending their style, motion, and mood cues into one output.

How many references can I provide?

Up to 30 images, 10 video clips, and 10 audio files, via `images_list`, `videos_list`, and `audios_list` respectively.

Why does this cost more than plain text/image-to-video?

Omni Reference uses the same per-second rate as the base model at each resolution ($0.374/sec at 720p, $0.187/sec at 480p, $0.935/sec at 1080p, $1.87/sec at 4K), but is billed on total video duration — the generated output plus every reference video you supply — not just the output length.

What resolution does it output?

720p by default. Pass a `resolution` value of `480p`, `720p`, `1080p`, or `4k` on the same endpoint to select a different output tier — price scales with the resolution chosen.

Does it support 1080p?

Yes — set `resolution` to `1080p` (natively rendered) or `4k` on this same endpoint. Both are priced above the 720p tier. The legacy dedicated [1080p](/playground/seedance-2.5-intl-omni-reference-1080p) and [4K](/playground/seedance-2.5-intl-omni-reference-4k) endpoints still work too, for existing integrations.

How long can generated videos be?

Between 4 and 30 seconds, configurable via the `duration` parameter.

Do I need to provide all three reference types?

No — provide any combination of images, videos, and audio. A text prompt alone still works, but references help steer style, motion, and mood more precisely.

Does this endpoint support Draft Mode?

Yes — set draft: true for a cheap 480p preview ($0.187/sec), then finalize the one you keep at 1080p by calling this same endpoint again with draft_request_id set to that draft's request_id, instead of a new prompt. See the [Draft Mode API guide](/seedance-2.5-draft-mode) for the full workflow.