Create video

OpenAI Sora video generation API in a compatible format. Supports text-to-video and image/video reference generation modes.

• POST /v1/videos: Create video task
• GET /v1/videos/{video_id}: Query task status
• GET /v1/videos/{video_id}/content: Retrieve the video binary content

Currently not available externally:

• GET /v1/videos
• POST /v1/videos/{video_id}/remix

Standard authentication headers:

• Content-Type: application/json

1. Use POST /v1/videos to create a task and obtain id
2. Poll GET /v1/videos/{id} until completed or failed
3. After completion, prefer using video_url returned in the response
4. For unified download, call GET /v1/videos/{id}/content

Core parameters:

• model: string: Required, public model name
• prompt: string: Recommended, video generation description
• seconds: number|string: Optional, target duration
• duration: number|string: Optional, duration alias
• size: string: Optional, output resolution
• mode: string: Optional, common values: t2v i2v i2v_first_last reference_material

Common compatibility parameters:

• aspect_ratio
• ratio
• quality
• resolution
• fps
• image
• image_url
• image_urls
• images
• reference_images
• input_reference
• end_image_url
• last_image_url
• video_urls
• audio_urls
• function_mode
• content
• callback_url
• external_task_id

Kling common extended parameters:

• model_name
• negative_prompt
• cfg_scale
• sound
• camera_control
• image_list
• video_id
• task_id
• watermark_info

4.1 Veo series

Note: stable, official, and similar labels indicate different groups, not external model suffixes; when calling, keep model as the base model name, and let the account or token group determine which group is actually used.

Veo 3.x main model

• veo3
• veo3-fast
• veo3-fast-frames
• veo3-frames
• veo3-pro
• veo3-pro-frames
• veo3.1
• veo3.1-fast
• veo3.1-pro
• veo3.1-components
• veo3.1-4k
• veo3.1-pro-4k

Recommendations:

• For text-to-video only, prefer the veo3* / veo3.1* base models
• For image-to-video or reference-to-video scenarios, it is recommended to explicitly pass mode + image_url/reference_images

4.2 Sora series

Base models:

• sora-2
• sora-2-pro

Recommendations:

• sora-2* is suitable for general-purpose use
• To use groups such as stable / official, keep model as the base model name and do not append any suffix to the model name.

Request example:

{
  "size": "1280x720",
  "images": [
    "https://example.png"
  ],
  "model": "sora-2",
  "prompt": "this is an example",
  "duration": 12,
  "fps": "24",
  "seed": "20231024",
  "mode": "i2v"
}

4.3 HappyHorse series

Public model:

• happyhorse-1.0

Recommendations:

• happyhorse-1.0 are distinguished by mode into t2v, i2v, r2v, and video_edit.
• In Sora-compatible format, it is recommended to pass media items in the content array for reference images.
• Use type=text for text items and type=image_url for media items, and optionally use role and name to indicate the asset’s purpose.

Request example:

{
  "size": "720x1280",
  "model": "happyhorse-1.0",
  "mode": "r2v",
  "prompt": "this is an example",
  "duration": 15,
  "content": [
    {
      "type": "text",
      "text": "this is an example"
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.png"
      },
      "role": "reference_image",
      "name": "image1"
    }
  ],
  "parameters": {
    "resolution": "720P",
    "ratio": "9:16",
    "duration": 15,
    "seed": 1
  }
}

4.4 Seedance Series

Public model:

• doubao-seedance-1-0-lite-t2v-250428
• doubao-seedance-1-0-lite-i2v-250428
• doubao-seedance-1-0-pro-250528
• doubao-seedance-1-0-pro-fast-251015
• doubao-seedance-1-5-pro-251215
• doubao-seedance-2-0-260128
• doubao-seedance-2-0-fast-260128

Recommendations:

• For text-to-video, prefer *-t2v-* or pro / fast
• For image-to-video, prefer *-i2v-*
• For scenarios with complex reference materials, prefer content

Request example:

{
  "size": "1280x720",
  "model": "doubao-seedance-2-0-fast-260128",
  "prompt": "this is an example",
  "duration": 15,
  "fps": "24",
  "seed": "20231024",
  "mode": "reference_material",
  "content": [
    {
      "type": "text",
      "text": "this is an example"
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.png"
      },
      "role": "reference_image",
      "name": "image1"
    },
    {
      "type": "audio_url",
      "audio_url": {
        "url": "https://example.mp3"
      },
      "role": "reference_audio",
      "name": "audio1"
    }
  ]
}

4.5 Grok Video Series

Public model:

• grok-imagine-1.0-video
• grok-imagine-video-1.5-preview
• grok-video-3

Common parameters:

• prompt
• ratio / aspect_ratio
• resolution / size
• seconds / duration
• image / image_url / input_reference
• reference_images

Example:

{
  "model": "grok-imagine-1.0-video",
  "prompt": "雨夜霓虹街道上的电影感推镜，光影丰富，运动自然",
  "reference_images": [
    "https://example.com/ref-1.jpg"
  ],
  "seconds": 10,
  "aspect_ratio": "16:9",
  "resolution": "720P"
}

4.6 Kling Video Main Model

Public base model:

• kling-video

Required:

• model
• model_name

Supported model_name:

• kling-v1
• kling-v1-5
• kling-v1-6
• kling-v2-master
• kling-v2-1
• kling-v2-1-master
• kling-v2-5-turbo
• kling-v2-6
• kling-v3

Common mode:

• t2v
• i2v
• multi_i2v
• extend

Minimum input parameters:

• Text-to-video: model + model_name + prompt + mode=t2v
• Image-to-video: model + model_name + prompt + mode=i2v + image
• Multi-image reference: model + model_name + prompt + mode=multi_i2v + image_list
• Video extension: model + model_name + mode=extend + video_id

Example:

{
  "model": "kling-video",
  "model_name": "kling-v2-6",
  "mode": "t2v",
  "prompt": "海边日落镜头，电影感，风吹长发",
  "duration": 5,
  "aspect_ratio": "16:9"
}

4.7 MiniMax / Hailuo Video Series

Public model:

• minimax-h3
• hailuo-2.3

Recommendations:

• minimax-h3 supports text-to-video, image references, and audio references. We recommend mixing text and assets via content; you can also use compatible fields such as image_urls and audio_urls.
• hailuo-2.3 requires a first-frame image. We recommend passing the image_url field of role=first_frame in content, or using image_urls to pass the first-frame image.
• minimax-h3 commonly uses duration=5~15 and resolution=1440p; hailuo-2.3 commonly uses duration=6 or 10 and resolution=768p.

minimax-h3 Example:

{
  "model": "minimax-h3",
  "prompt": "A cinematic product video with smooth camera motion",
  "content": [
    {
      "type": "text",
      "text": "Use the product image as reference and match the rhythm of the audio."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/product.jpg"
      },
      "role": "reference_image",
      "name": "product"
    },
    {
      "type": "audio_url",
      "audio_url": {
        "url": "https://example.com/music.mp3"
      },
      "role": "reference_audio",
      "name": "music"
    }
  ],
  "duration": 8,
  "resolution": "1440p",
  "aspect_ratio": "16:9",
  "generate_audio": true
}

hailuo-2.3 Example:

{
  "model": "hailuo-2.3",
  "prompt": "The camera slowly pushes in, natural motion and cinematic lighting.",
  "content": [
    {
      "type": "text",
      "text": "The camera slowly pushes in, natural motion and cinematic lighting."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/first-frame.jpg"
      },
      "role": "first_frame",
      "name": "first"
    }
  ],
  "duration": 10,
  "resolution": "768p",
  "aspect_ratio": "16:9"
}

• mode=t2v

• Minimum input: model + prompt

• mode=i2v

• Minimum input: model + prompt + image_url

• mode=i2v_first_last

• Minimum input: model + prompt + image_url + end_image_url

• mode=reference_images

• Minimum input: model + prompt + reference_images

• mode=reference_material

• Minimum input: model + prompt + (image_urls/video_urls/audio_urls 至少一种)

Gemini Omni usage instructions

• gemini-omni is the publicly exposed video model name of new-api, and can be called directly via POST /v1/videos.
• mode=t2v: text-to-video, with the minimum input parameter being model + prompt.
• mode=r2v: reference image/reference material generation; images can be placed in image, image_url, images, image_urls, reference_images, input_reference, or content.
• mode=edit: video editing; videos can be placed in video, video_url, videos, or content, and reference images can also be provided at the same time.
• The duration field can use seconds or duration, and will be automatically mapped to the 4 / 6 / 8 / 10 second options.

Kling 4.0

Public models: kling-4.0 (720p) and kling-4.0-1080p. The native format supports POST /text-to-video/kling-4.0, POST /image-to-video/kling-4.0, and POST /omni-video/kling-4.0, with queries made through GET /tasks?task_ids=任务ID. Native requests do not require model; resolution is selected by settings.resolution, with 1080p as the default. Native reference image/video types are refer_image and feature_video; creation returns a data object, while queries return a data array. Requests can be submitted through POST /api/v3/contents/generations/tasks or POST /v1/videos and queried through GET /{task_id} on the corresponding path. OpenAI/Sora directly use prompt, seconds, size, and input_reference; Seedance directly uses content and top-level duration/resolution/ratio/mode. Neither requires additional contents/settings/options. The legacy envelope remains supported: in contents, type=prompt is text; type=reference_image and reference_video are reference materials; and first_frame and last_frame are the first and last frames. kling-4.0 corresponds to 720p, and kling-4.0-1080p corresponds to 1080p; ensure that settings.resolution matches the model name. In settings, use duration (4–30 seconds), resolution (720p/1080p), and aspect_ratio (21:9, 16:9, 9:16, 1:1). audio is available only for all-purpose mode with standard reference materials; omit the audio switch for text-to-video and first/last-frame modes. The current API supports up to 10 reference images and 5 reference videos, and does not support reference audio. The legacy kling-video and model_name modes remain available as backward-compatible interfaces.

{
  "model": "kling-4.0",
  "prompt": "A cinematic product shot",
  "seconds": "4",
  "size": "1280x720",
  "input_reference": "https://example.com/product.jpg"
}

Kling 4.0 can directly use the Seedance format with content/duration/resolution/ratio/mode, or the OpenAI video format with prompt/seconds/size/input_reference; neither requires additional contents/settings/options. The legacy contents/settings/options envelope is retained for compatibility only. Available mode values are t2v (text only), i2v (a single first frame), i2v_first_last (first and last frames), reference_images (standard reference images), and reference_material (image or video references). If mode is omitted and no tool is explicitly specified, reference_material is used by default when reference materials are provided; otherwise, t2v is used by default. Reference modes use standard reference-image roles and do not silently convert explicitly specified first and last frames into reference images. For first and last frames, explicitly select i2v or i2v_first_last and omit the aspect-ratio parameter. Reference audio is not currently supported. settings.audio is available only in all-purpose mode for reference images/reference materials; omit the audio switch for t2v, i2v, and i2v_first_last.

Vidu Q4

Vidu Q4: vidu-q4 is fixed at 720p, and vidu-q4-540p is fixed at 540p. Supports generating videos from 3 to 16 seconds. At least 1 image is required, with up to 15 reference images supported. Reference generation can include up to 3 MP3 audio clips, each 3 to 12 seconds long and up to 50 MB. Pure text-to-video, specified first and last frames, and reference videos are not supported. Multiple images are used as joint references and do not indicate first or last frame positions. Tasks can be created via POST /v1/videos, POST /api/v3/contents/generations/tasks, or the Vidu formats POST /ent/v2/img2video and POST /ent/v2/reference2video. For the unified endpoint, mode=reference_images is recommended; use mode=reference_material when including audio. Reference audio can use audio_urls. The Seedance format can also use the audio_url field of the standard content; the audio Boolean field controls audio generation and is used separately from the reference audio URL list. The resolution must match the model name.