Skip to content

Seedream โ€‹

Updated

Endpoint: POST https://vod.intl.tencentcloudapi.comAction: CreateAigcImageTask

Seedream is an image generation model that supports text-to-image and multi-image reference generation (up to 10 reference images on 5.0-pro). Versions 5.5, 5.0-flash, 5.0-pro, 5.0-lite, 4.5, and 4.0 are available. The latest 5.5 (launched 2026-11-17) shares the same parameters as 5.0-flash (launched 2026-11-12), and both support image layer decomposition.

See Version differences for specs and Calling by version for request examples.

Basic Information โ€‹

ItemValue
ModelNameSI (5.5 / 5.0-flash / 5.0-pro) / Seedream (5.0-lite / 4.5 / 4.0)
ModelVersion5.5 / 5.0-flash / 5.0-pro / 5.0-lite / 4.5 / 4.0
Existing Seedream defaultModelVersion=4.0 / AspectRatio=1:1
Guardrail disable supportNot supported

Version differences โ€‹

VersionResolutionAspect ratioReference images
5.5 / 5.0-flash1K / 1.5K / 2K / 4K (OutputConfig.Resolution)1:1 / 16:9 / 9:16 / 4:3 / 3:4 / 3:2 / 2:3 / 21:9Supported (image-to-image, layer decomposition)
5.0-pro1K / 1.5K / 2K pixel specification1:1 / 16:9 / 9:16 / 4:3 / 3:4Up to 10
5.0-lite / 4.5 / 4.0Preset / pixel specification1:1 / 16:9 / 9:16 / 4:3 / 3:4Reference images supported
Capability5.55.0-flash5.0-pro5.0-lite / 4.5 / 4.0
ModelNameSISISISeedream
Text to imageโœ…โœ…โœ…โœ…
Reference image to imageโœ…โœ…โœ… (up to 10)โœ…
Layer decomposition (layer_decomposition)โœ…โœ…โ€”โ€”
How resolution is setResolution takes 1K / 1.5K / 2K / 4K directlySame as 5.5Resolution + pixel ExtInfo.sizePreset / pixel specification

The 5.5 / 5.0-flash aspect ratios are compiled from the integration guide; the guide writes "3:3/2:3", which from context should be 3:2 / 2:3. The guide documents both versions in one parameter table ("SI 5.0-flash/5.5") and does not describe differences in quality or speed.

The pixel size of 5.0-pro is specified in ExtInfo.AdditionalParameters.size. On a square basis, it is 1024x1024 / 1536x1536 / 2048x2048. 1.5K is not put in OutputConfig.Resolution; specify it with size.

Request Parameters โ€‹

ParameterRequiredTypeDescription
ModelNameRequiredString5.5 / 5.0-flash / 5.0-pro: SI. Existing 5.0-lite / 4.5 / 4.0: Seedream
ModelVersionRequiredString5.5 / 5.0-flash / 5.0-pro / 5.0-lite / 4.5 / 4.0
PromptRequiredStringGeneration prompt
FileInfos.NOptionalArrayReference image array. 5.0-pro is up to 10. Specify Type=Url, Url in each item
OutputConfig.AspectRatioOptionalString1:1 / 16:9 / 9:16 / 4:3 / 3:4, etc.
OutputConfig.OutputImageCountOptionalIntegerNumber of images to generate. The example below generates 1 images
OutputConfig.ResolutionOptionalStringResolution preset. Omit when specifying pixels with 1.5K
OutputConfig.StorageModeOptionalStringTemporary / Permanent
ExtInfoOptionalStringExtension options. In AdditionalParameters, put a JSON string, then encode the whole again as a JSON string. 5.0-pro uses size; 5.5 / 5.0-flash layer decomposition uses layer_decomposition

Calling by version โ€‹

VersionTextImageLayer decompositionExamples
5.5โœ…โœ…โœ…5.5
5.0-flashโœ…โœ…โœ…5.0 Flash
5.0-proโœ…โœ…โ€”5.0 Pro
5.0-lite / 4.5 / 4.0โœ…โœ…โ€”Existing versions

5.5 โ€‹

Launched 2026-11-17. ModelName=SI, ModelVersion=5.5. Parameters are identical to 5.0 Flash: pass 1K / 1.5K / 2K / 4K directly in OutputConfig.Resolution, and layer decomposition also uses layer_decomposition. To migrate from 5.0-flash, just change ModelVersion to 5.5.

Text to image

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.5",
  "Prompt": "A young artist in a bunny cosplay outfit hand-copies the image on her computer screen onto a canvas as an oil painting",
  "OutputConfig": {
    "StorageMode": "Temporary",
    "Resolution": "2K",
    "AspectRatio": "16:9"
  }
}

Image to image: put the reference image in FileInfos.

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.5",
  "FileInfos": [
    {
      "Type": "Url",
      "Url": "https://<host>/reference.png"
    }
  ],
  "Prompt": "Keep the room's structure unchanged; restyle the furniture and soft furnishings in a Japanese wabi-sabi style",
  "OutputConfig": {
    "StorageMode": "Temporary",
    "Resolution": "1K",
    "AspectRatio": "9:16"
  }
}

Layer decomposition: see 5.0 Flash / Layer decomposition and change ModelVersion to 5.5.

5.0 Flash โ€‹

Launched 2026-11-12. ModelName=SI, ModelVersion=5.0-flash. Pass 1K / 1.5K / 2K / 4K directly in OutputConfig.Resolution โ€” no pixel size in ExtInfo as on 5.0-pro.

Text to image

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.0-flash",
  "Prompt": "A young artist in a bunny cosplay outfit hand-copies the image on her computer screen onto a canvas as an oil painting โ€” and the photo itself captures her painting this recursive image",
  "OutputConfig": {
    "StorageMode": "Temporary",
    "Resolution": "1K",
    "AspectRatio": "9:16"
  }
}

Image to image: put the reference image in FileInfos.

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.0-flash",
  "FileInfos": [
    {
      "Type": "Url",
      "Url": "https://<host>/reference.png"
    }
  ],
  "Prompt": "Keep the room's structure unchanged; restyle the furniture and soft furnishings in a Japanese wabi-sabi style",
  "OutputConfig": {
    "StorageMode": "Temporary",
    "Resolution": "1K",
    "AspectRatio": "9:16"
  }
}

5.0 Flash / Layer decomposition โ€‹

Splits the input image into multiple layers. Set layer_decomposition=true in ExtInfo.AdditionalParameters; Prompt can be omitted.

Parameter (inside AdditionalParameters)TypeDescription
layer_decompositionBooleantrue: decompose into layers; false: no decomposition
json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.0-flash",
  "FileInfos": [
    {
      "Type": "Url",
      "Url": "https://<host>/reference.png"
    }
  ],
  "OutputConfig": {
    "StorageMode": "Temporary"
  },
  "ExtInfo": "{\"AdditionalParameters\":\"{\\\"layer_decomposition\\\":true}\"}"
}

Decoded AdditionalParameters:

json
{ "layer_decomposition": true }

Mind the two levels of escaping in ExtInfo

ExtInfo is a string, and AdditionalParameters inside it is another string, so the inner quotes must be written as \\\". The example in the integration guide has incomplete escaping plus a trailing comma and a full-width comma, so copying it as-is fails to parse โ€” use the form on this page. The guide does not describe the layered output structure (layer count, how the per-layer FileInfos are organized); rely on the actual DescribeTaskDetail response.

Request examples: 5.0 Pro and existing versions โ€‹

5.0 Pro / Text-based generation โ€‹

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.0-pro",
  "Prompt": "A blue ceramic cup beside an orange on a white table. Product photography, soft daylight.",
  "OutputConfig": {
    "StorageMode": "Temporary",
    "OutputImageCount": 1,
    "Resolution": "2K",
    "AspectRatio": "1:1"
  },
  "ExtInfo": "{\"AdditionalParameters\":\"{\\\"size\\\":\\\"2048x2048\\\"}\"}"
}

5.0 Pro / Reference and pixel specification โ€‹

Pass the reference images in order into the FileInfos array. The request below generates one image of size 1536x1536.

json
{
  "SubAppId": 123456789,
  "ModelName": "SI",
  "ModelVersion": "5.0-pro",
  "Prompt": "Keep the blue cup in the reference images. Place a small orange next to it on a white table. Product photography.",
  "FileInfos": [
    {
      "Type": "Url",
      "Url": "https://mpstestmodel-1315536146.cos.ap-singapore.myqcloud.com/justin/wiki-examples/reference-cup.png"
    }
  ],
  "OutputConfig": {
    "AspectRatio": "1:1",
    "OutputImageCount": 1,
    "StorageMode": "Temporary"
  },
  "ExtInfo": "{\"AdditionalParameters\":\"{\\\"size\\\":\\\"1536x1536\\\"}\"}"
}

Existing version / Text-based generation โ€‹

json
{
  "SubAppId": 123456789,
  "ModelName": "Seedream",
  "ModelVersion": "4.0",
  "Prompt": "a serene japanese garden, soft morning light",
  "OutputConfig": {
    "AspectRatio": "1:1",
    "OutputImageCount": 1,
    "StorageMode": "Temporary"
  },
  "SessionContext": "job-001"
}

Existing version / Reference image request โ€‹

json
{
  "SubAppId": 123456789,
  "ModelName": "Seedream",
  "ModelVersion": "4.0",
  "Prompt": "Change the background to a quiet garden; preserve the subject.",
  "OutputConfig": {
    "AspectRatio": "1:1",
    "OutputImageCount": 1,
    "StorageMode": "Temporary"
  },
  "SessionContext": "job-001",
  "FileInfos": [
    {
      "Type": "Url",
      "Category": "Image",
      "Url": "https://<host>/reference.jpg"
    }
  ]
}

Response Example โ€‹

Query the TaskId received from the generation request with DescribeTaskDetail. When AigcImageTask.Status=FINISH, ErrCode=0, read the result from Output.FileInfos. Below is the completion response of a 2048x2048 text generation request.

json
{
  "AigcImageTask": {
    "Status": "FINISH",
    "ErrCode": 0,
    "Progress": 100,
    "Output": {
      "FileInfos": [
        {
          "FileUrl": "https://<output-host>/<image>.jpg",
          "MetaData": {
            "Width": 2048,
            "Height": 2048,
            "Container": "image2",
            "Size": 231055
          }
        }
      ]
    }
  }
}

Special Settings โ€‹

Model Call Name and Pixel Specification โ€‹

5.5 / 5.0-flash / 5.0-pro use ModelName=SI in VOD; 5.0-pro uses ModelName=Seedream in MPS. Distinguish the model call name and extension parameter format of the two paths.

For reference generation, in FileInfos put Type=Url, Url. 5.0-pro can pass up to 10 images. 1536x1536 output omits Resolution and is specified with ExtInfo's size.

Built with VitePress ยท Deployed on Tencent EdgeOne Pages