Seedance 2.5 Image-to-Video API

Overview

Seedance 2.5 Image-to-Video is the image-to-video interface of ByteDance's Seedance 2.5 model on iCreat: input a text prompt plus reference images to generate video clips with natively synchronized audio. The full model supports up to 50 multimodal reference inputs; the current iCreat interface exposes text + image references (1-30 images). Developers invoke bytedance/seedance-2-5/image-to-video through an asynchronous task queue, with 480p / 720p output and 4-30 second durations.

bytedance/seedance-2-5/image-to-video

Core Capabilities

Reference-image-to-video. Use 1-30 reference images as visual anchors (role: reference_image) with a text prompt to drive video generation—ideal for animating product shots, character art, and scene references.

30-second one-take generation. A full narrative arc—setup, progression, turning point, payoff—fits in a single render without post-stitching.

Native synchronized audio. Dialogue, sound effects, and music are generated together with the picture in one pass (generate_audio: true).

Multilingual prompts. Supports Chinese, English, Japanese, Indonesian, Spanish, and Portuguese.

Asynchronous task design. Submit → poll → retrieve workflow that is batch-friendly and easy to integrate with fault tolerance.

Review mechanism. need_review submits reference images containing faces or copyrighted IP for official review to avoid generation failure.

Model Comparison

Seedance 2.5 Image-to-Video (iCreat) vs Seedance 2.5 (Full Model)

Seedance 2.5 Image-to-Video vs Gemini Omni Flash vs Kling V3 Omni

Input

Text prompt is required. Reference images are passed through the multimodal content array.

Reference image requirements:

• Formats: jpeg、png、webp、bmp、tiff、gif、heic、heif
• Aspect ratio (w/h): 0.4–2.5
• Width and height: 300–6000 px
• Max 30 MB per image; max 64 MB request body

Parameters

Chinese prompts must not exceed 2000 characters; English prompts must not exceed 2000 words. Overly long prompts may dilute key information and cause the model to ignore details.

Pricing

Total cost = unit price × output video duration

Quick Start

Get your API Key from https://icreat.ai/hub/keys.

Submit a reference-image-to-video task:

curl --fail-with-body --connect-timeout 10 --max-time 60 \
  -X POST https://api.icreat.ai/v1/task/submit/bytedance/seedance-2-5/image-to-video \
  -H "Authorization: Bearer ${ICREAT_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
  "content": [
    {
      "type": "text",
      "text": "A cat is playing with a ball"
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/reference.jpg"
      },
      "role": "reference_image"
    }
  ],
  "generate_audio": true,
  "ratio": "3:4",
  "duration": 5,
  "resolution": "480p"
}'

The server returns:

{ "task_id": "task-xxx" }

Poll task status:

curl --connect-timeout 10 --max-time 60 \
  -X POST "https://api.icreat.ai/v1/task/query-status" \
  -H "Authorization: Bearer ${ICREAT_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
  "task_id": "task-xxx"
}'

When the status is SUCCEEDED, fetch the result:

curl --connect-timeout 10 --max-time 60 \
  -X POST "https://api.icreat.ai/v1/task/get-result" \
  -H "Authorization: Bearer ${ICREAT_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
  "task_id": "task-xxx"
}'

Output

submit → task_id → query-status → get-result

After submitting, poll query-status with the task_id until the status is SUCCEEDED, then call get-result with the same task_id to retrieve the generated video.

Use Cases

Product image animation. Turn static product shots into in-use scenarios, feature demos, and selling-point videos with synchronized audio—ideal for e-commerce marketing.

Character image animation. Animate character reference images into action clips, using need_review for face review—suited to short drama and IP content.

Brand films and ads. A 30-second spot is one render; feed product shots and brand references so the look holds from the first frame to the last.

Dialogue-led scenes. Quoted spoken lines and described sound design keep performance, timing, and effects aligned.

Limitations

The current iCreat interface accepts text + image references only; video and audio reference inputs are not exposed (the full model supports up to 50 multimodal inputs).

Chinese prompts must not exceed 2000 characters; English prompts must not exceed 2000 words. Overly long prompts may dilute key information.

Reference images containing faces or copyrighted IP require need_review: true; otherwise generation may fail.

API output resolution is capped at 480p / 720p.

FAQ

How do I get an API Key?

Create one at https://icreat.ai/hub/keys and pass it via the Authorization: Bearer header.

Why does my task fail when the reference image contains a face?

Set need_review: true for faces or copyrighted IP. Setting it to false for such images may cause failure; setting it to true for images that do not need review only extends generation time.

What does duration: -1 mean?

It lets the model choose the video length automatically instead of a fixed 4-30 second value.

What does ratio: adaptive do?

The model automatically selects the aspect ratio based on the input image.

480p or 720p?

Use 480p for fast, low-cost iteration and 720p when more detail is required.

What is the difference between the Image-to-Video interface and the full Seedance 2.5 model?

iCreat currently exposes text + reference-image input only (image-to-video); the full model also supports video references, audio references, timestamp-level editing, and video extension.