You need to enable JavaScript to run this app.
Lake AI Service

Lake AI Service

Copy page
Download PDF
Video understanding
Enhanced video content understanding (Doubao series)
Copy page
Download PDF
Enhanced video content understanding (Doubao series)

Operator introduction

Description

The video content understanding operator supports using the Doubao model to understand video files, including parsing video content and generating natural language descriptions.

Limitations

  • The storage space occupied by video files must be less than 1 GiB.
  • Currently, audio information in video files is not supported for understanding.
  • Video links must be accessible via the public network or Volcano Engine intranet.

Key features

  • Compress the video to within 50 MB, then use the Doubao model for video understanding.
  • Supported video formats: mp4, wmv, webm, mkv, m4v, flv, avi, mov. Due to the wide variety of video file format variants, not all files can be guaranteed to be recognized. Please verify through testing that files can be properly recognized.

Performance description

  • In theory, the larger the video, the longer the processing time required.
  • Actual processing time is affected by factors such as video quality and network environment. It is recommended to use in the Volcano Engine intranet environment.

Caution and prerequisites

Details

Caution and prerequisites

Costs

Before calling an operator, you need to understand the model invocation costs associated with using the operator. For details, see Large model invocation billing.

Authentication (API Key)

Before calling an operator, you need to generate an API Key for operator invocation. It is recommended to configure the API Key as an environment variable to ensure safer operator calls. For details, see Obtain and configure API Key.

BaseURL

Before calling an operator, you need to determine the BaseURL for operator invocation based on the region where your current LAS service is deployed. This is used to configure the path parameter values for operator calls.
For details, see Obtain the Base URL. The Examples below are for reference only; when making actual calls, replace the path values with those corresponding to your region.

Billing
  • Billing standards

    Sub-item
    Billing standard description

    Billing item

    There are two billing items: Inference service (input) and Inference service (output). The total fee payable is the sum of all billing items.

    Billing type

    Pay-as-you-go, unit: CNY per million tokens, billed hourly based on actual usage.

    Unit price

    • The unit price depends on the type of video link you input. The unit prices for public network links and VolcanoEngineTOS path links are different.
    • For different input video link scenarios, the unit price of each billing item is tiered based on the model input length at the time of model invocation. The unit price varies depending on the input length tier.

    warning

    • When the input video link is a TOS path, the unit price is usually lower.
  • Billing details
    Billing formula: Total cost = Total price of billing item 1 + Total price of billing item 2 = (Unit price of billing item 1 * usage) + (Unit price of billing item 2 * usage)

    Model type
    Sub-scenario
    (Input video link type)
    Condition
    (thousand tokens)
    Unit price: Billing item 1
    Inference service (input)
    Unit price: Billing item 2
    Inference service (output)
    • doubao-seed-1.6
    • doubao-seed-1.8

    Video link: Public network link

    Input length [0, 128]

    4.7 CNY per million tokens

    20 CNY per million tokens

    Input length (128, 256]

    5.9 CNY per million tokens

    30 CNY per million tokens

    Video link: TOS path

    Input length [0, 128]

    2.6 CNY per million tokens

    20 CNY per million tokens

    Input length (128, 256]

    3.8 CNY per million tokens

    30 CNY per million tokens

    • doubao-seed-2.0-pro

    Video link: Public network link

    Input length [0, 32]

    8.6 CNY per million tokens

    28.9 CNY per million tokens

    Input length (32, 128]

    11.3 CNY per million tokens

    42.4 CNY per million tokens

    Input length (128, 256]

    19.3 CNY per million tokens

    82.8 CNY per million tokens

    Video link: TOS path

    Input length [0, 32]

    6.6 CNY per million tokens

    27 CNY per million tokens

    Input length (32, 128]

    9.3 CNY per million Tokens

    40.4 CNY per million Tokens

    Input length (128, 256]

    17.4 CNY per million Tokens

    80.8 CNY per million Tokens

  • Billing estimation

    • Based on the above billing details table, you can estimate approximately how many tokens the input video contains. Combined with the operator request code, you can estimate the overall model input length in tokens.
    • Caution: This estimation result cannot be used as the exact input length data and is for reference only. When processing tasks, there will also be system prompts and other inputs, but these other inputs are generally short.

Online experience

LAS provides you with the capability for "online experience" and offers a certain amount of free usage quota. No configuration is required to experience the data processing effect of the LAS operator online.

warning

The current operator's online experience allows free usage of up to 500,000 tokens for model consumption. Any excess will be billed according to the operator's billing metrics. For details on billing metrics and logic for each operator, please refer to Large model invocation billing.

Online experience entry

After logging in and entering LAS console , locate the current operator card, hover the mouse over the operator card, and click the "online experience" button.
Image

Online experience operation demonstration

  • LAS provides you with multiple sample video files. You can also delete sample files and manually upload files.
  • During the online experience, you can flexibly set the operator's processing parameters.

API invocation

Interface description

Invoke the las_vlm_video operator for video understanding via the Chat interface.

Request parameters

Parameter name
Type
Required
Example
Description
las_extra_params
las_extra_params
Yes
Additional operator-specific parameters.
model
string
Yes
doubao-seed-1-8-251228
Specifies the model used for video understanding. Supported model names:
  • doubao-seed-1-8-251228
  • doubao-seed-2-0-pro-260215
messages
list of chat_message
Yes
A list of messages that make up the conversation so far. Different models support different message types, such as text, images, and videos. For usage of this field, refer to the messages parameter in the request body of the Chat API. Video URLs support public HTTP links and TOS paths.
max_tokens
integer
No
10000
Value range: varies by model. For details, see Model List.
Maximum length of the model response (unit: token).
  • The model response does not include reasoning content. Model response = model output - model reasoning content (if any).
  • The total length of output tokens is also limited by the model context length.
max_completion_tokens
integer
No
10000
> For models that support this field and usage instructions, see Documentation.

Value range: [0, 64k].
Controls the maximum length of the model output, including both the model response and reasoning content (unit: token). After this parameter is configured, the model can output ultra-long content, max_tokens (default value: 4k) becomes invalid, and the model outputs content (response and reasoning content) as needed until it reaches the value configured for max_completion_tokens.
Cannot be set together with max_tokens; otherwise, an error is returned.
stop
list of string
No
Default value: null.
The model stops generating when it encounters any string specified in the stop field, and the string itself is not output. Up to 4 strings are supported.
> Models with deep thinking capabilities do not support this field.
frequency_penalty
float
No
1.0
Value range: [-2.0, 2.0]. Frequency penalty coefficient. If the value is positive, new tokens are penalized based on their frequency in the text, reducing the likelihood of verbatim repetition.
presence_penalty
float
No
1.0
Value range: [-2.0, 2.0]. Presence penalty coefficient. If the value is positive, new tokens are penalized based on whether they have appeared in the text so far, increasing the likelihood that the model discusses new topics.
temperature
float
No
1
Value range: [0, 2]. Sampling temperature. Controls how much smoothing is applied to the probability distribution of each candidate token during text generation. When set to 0, the model only considers the token with the highest log probability. Higher values, such as 0.8, make the output more random; lower values, such as 0.2, make the output more focused and deterministic. It is generally recommended to adjust only one of temperature or top_p, not both.
top_p
float
No
0.5
Value range: [0, 1]. Nucleus sampling probability threshold. The model considers token results whose probability mass falls within top_p. When set to 0, the model only considers the token with the highest log probability. A value of 0.1 means only the top 10% tokens by probability mass are considered. Larger values increase randomness, while smaller values increase determinism. It is generally recommended to adjust only one of temperature or top_p, not both.
logprobs
boolean
No
true
> Models with deep thinking capabilities do not support this field.

Whether to return log probabilities of output tokens:
  • false: Do not return log probability information.
  • true: Return the log probability of each output token in the message content.
top_logprobs
integer
No
10
> Models with deep thinking capabilities do not support this field.

Value range: [0, 20]. Specifies the number of most likely tokens to return at each output token position, each associated with a log probability. The top_logprobs parameter can be set only when logprobs is true.
logit_bias
map[str,int]
No
> Models with deep thinking capabilities do not support this field.

Adjusts the probability that specified tokens appear in the model output, making the generated content better align with specific preferences. The logit_bias field accepts a map where each key is a token ID in the vocabulary (obtained by using the tokenization interface), and each value is the bias value for that token. Value range: [-100, 100]. -1 reduces the likelihood of selection, 1 increases the likelihood, -100 completely prohibits selecting the token, and 100 makes only that token selectable. The actual effect may vary by model.
thinking
request_thinking
No
Controls whether deep thinking mode is enabled for the model. Deep thinking mode is enabled by default and can be manually disabled. For usage of this field, refer to the thinking parameter in the request body of the Chat API.
response_format
request_response_format
No
The model output content must follow the format specified here. For usage of this field, refer to the response_format parameter in the request body of the Chat API.
tools
list of chat_tool
No
List of tools to be called, which may be included in the model response. Configure this structure when you need the model to return tools to be called. For usage of this field, refer to the tools parameter in the request body of the Chat API.
tool_choice
string or object
No
> Only doubao-seed-1-6-* and later-generation models support this field.Specifies whether the model response for this request contains tools to be called. When no tools are specified, none is the default value. If tools exist, auto is the default value.Tool selection mode (string)
  • none: The model response cannot contain tools to be called.
  • required: The model response must contain tools to be called. When selecting this option, confirm that suitable tools exist to reduce model hallucinations.
  • auto: The model independently determines whether to include tools to be called in the response.
Tool call (object)**
  • Specifies the scope of tools to be called. The model response may contain only the specified tool information. When selecting this option, confirm that the tool meets user needs to reduce model hallucinations. tool_choice.name specifies the name of the tool to be called, and tool_choice.type specifies the call type.

Response parameters

Parameter name
Type
Example
Description
id
string
0217426318107460cfa43dc3f3683b1de1c09624ff49085a456ac
Unique identifier of this request.
model
string
Name and version of the model actually used for this request.
created
integer
Unix timestamp, in seconds, when this request was created.
object
string
chat.completion
Fixed as chat.completion.
choices
list of choice
Model output content for this request.
usage
usage
Token usage for this request.
las_extra_result
las_extra_result
Additional operator-specific return attributes.

Examples

Request example

curl --location "https://operator.las.ap-southeast-1.volces.com/api/v1/chat/completions" \
--header "Content-Type: application/json" \
--header "Authorization: Bearer $LAS_API_KEY" \
--data '{
    "operator_id": "las_vlm_video",
    "operator_version": "v1",
    "data": {
       "las_extra_params": {
             "operator_id": "las_vlm_video",
             "operator_version": "v1"
             },
        "model_name":"doubao-seed-1-8-251228",
        "messages": [
            {
                "content": [
                    {
                        "video_url": {
                            "url": "https://las-ai-cn-beijing-baseline.tos-cn-beijing.volces.com/operator_cards_serving/public/baseline/las_vlm_video/v1/eating_56.mp4"
                        },
                        "type": "video_url"
                    },
                    {
                        "text": "",
                        "type": "text"
                    }
                ],

                "role": "user"
            }
        ]
    }
}'

Response example

{
    "metadata": {
        "task_status": "COMPLETED",
        "business_code": "0",
        "error_msg": "",
        "request_id": "7517a8d9fa003a570777e2a390977f73"
    },
    "data": {
        "vlm_result": {
            "id": "0217610362616819fc4dc47d27ebf7239e22958e08dac0b3c4dd9",
            "object": "chat.completion",
            "created": 1761036294,
            "model": "doubao-seed-1.8",
            "serviceTier": "default",
            "choices": [
                {
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "content": "The video unfolds with a brightly colored two-layer cake as the core, and the overall atmosphere is fantastical and with a sense of humor. Here is the specific analysis:\ n\ n\ n ###** 1. Scene and object details **\ n- ** Main cake **: The bottom layer is dark red, and the surface is covered with smooth red frosting; the top layer is light yellow, decorated with red wavy stripes, and the top is covered with thick red frosting. The frosting is stacked with beige cream flowers, and the top of the cream is inserted with a red cherry. The bottom edge of the cake is decorated with yellow stars, and the overall shape is exquisite and inviting.\ n- ** Background Environment **: The background is a blue carved window. The window frame lines are smooth, and the dark purple light is revealed, creating a mysterious atmosphere. The visible part on the left is a red bookshelf, with faint outlines of books; on the right hangs a painting of a hand holding a blue magic wand with pink stars twinkling at the end, suggesting that the scene may be related to "magic".\ n\ n ###** 2. Core characters and dynamics **\ n The key twist in the video is the green creature that suddenly appears at the top of the cake:\ n- ** Initial state **: The creature "emerges" from the cream of the top cake, initially showing only its head, eyes closed, mouth raised, revealing two sharp teeth, with a sly complacent expression (as if snickering or planning something). \ n- ** Expression Change **: Then the creature suddenly opens its eyes, the eyeballs instantly become big and round (highlighting surprise or excitement), the mouth opens, revealing more fangs, and the smile is exaggerated, changing from "sly" to "ecstasy" or "surprise", as if it has just finished a prank or discovered something interesting.\ n- ** Interactive Details **: When the creature appears, the cream flower is slightly pushed up but not scattered, and the red cherry at the top remains in place, suggesting that the creature is "drilled" from the inside of the cream, with light and fantastical movements.\ n\ n ###** 3. Ambience and underlying narrative **\ n- ** Fantastic tone **: The magic wand hanging on the right echoes the sudden appearance of the creature, suggesting that there may be a magical element in the scene - the cake may be enchanted, or the creature itself has magical powers that can be hidden in the food.\ n- ** Humor and mischievousness **: The expression of the green creature ranges from "bad laugh" to "exaggerated excitement", and the character image is mischievous and intelligent, contrasting with the delicate cake (the "sweetness" of the cake and the "cunning" of the creature), creating a relaxed and humorous atmosphere, possibly in the moment of a mischievous character's "surprise appearance". The video builds a short and vivid scene through vibrant color contrasts (red, yellow cakes and green creatures), fantastical elements (magic paintings, creatures that suddenly appear), and quick changes in the characters' expressions (sly → ecstatic): a mischievous green creature hidden in a cake suddenly appears in a magical environment, with the excitement of a successful mischief, and the overall delivery is relaxed and fantastical fun.",
                        "reasoning_content": "\ nThe user now needs me to analyze this video. First, I need to look carefully at the video content and describe all the elements I see. The video starts with a two-layer cake with a red bottom layer and a yellow top layer with red wavy stripes, topped with cream and cherries. The background is a blue window with a bookshelf on the left and a painting on the right with a magic wand and stars.\ n\ nThen, a green creature suddenly appears from the cream on the top of the cake. At first, it closes its eyes and shows a bad smile. Then the eyes open and become very large. The mouth opens and shows teeth. The expression changes from complacency to surprise or excitement? Need to pay attention to the process of biological change. \ n\ nNext, to analyze the atmosphere of the scene, the blue windows and bookshelves may be in a room, and the magic wand painting on the right may suggest that there are magic elements. The cake looks very tempting, and the green creature appears from the cake. It may appear suddenly, with a feeling of surprise or mischief.\ n\ nThe expression of the creature changes: first close the eyes and smile, then open the eyes, the eyes become larger, the mouth is open, and the teeth are exposed. It may be that you find yourself on the cake, or the reaction to the surrounding environment. You need to pay attention to details, such as the state of the cream, whether the cream changes when the creature appears, such as being pushed up.\ n\ nAlso consider the chronological order. The video is very short, about 3 seconds. The action ranges from static cake to creature appearance to expression change. The possible intention is to show a character who has suddenly appeared, with a humorous or fantastical color, because there is a background of magic painting, which may be related to magic, and the creature may have appeared through magic?\ n\ nSummary elements: cake (red bottom layer, yellow top layer, red frosting, cream, cherry), background (blue window, bookshelf, magic painting), green creature (expression change: closed eyes smirk → open eyes surprise/excitement), action (creature emerges from the top of the cake).\ n\ nPossible scene setting: a room with magic, the cake may be enchanted, or the creature is hidden in the cake and suddenly appears. The green of the creature contrasts with the bright color of the cake, highlighting the character. \ n\ nNeed to confirm if there are other details, such as the yellow decoration on the bottom of the cake, possibly stars or petals? There is a yellow shape on the bottom of the cake at the beginning of the video, which may be decoration. When the creature appears, the cherry on the top is still there, and the cream is jacked up, but the cherry is not dropped, indicating that the creature is drilled out of the cream.\ n\ nKey to the change of expression: When the eyes are closed, the corners of the mouth are raised, there is a smug smile, and it may be planning something; After opening the eyes, the eyes are round, the mouth is open, and the teeth are exposed. It may be surprised by their successful appearance, or the reaction after seeing the surrounding environment, they appear more lively.\ n\ nOverall atmosphere: Fantasy, humorous, with a little naughty feeling, because the creature's expression ranges from bad laugh to excitement, and may be a naugh"
                    },
                    "finish_reason": "stop"
                }
            ],
            "usage": {
                "prompt_tokens": 4903,
                "completion_tokens": 1133,
                "total_tokens": 6036,
                "prompt_tokens_details": {
                    "cached_tokens": 0
                },
                "completion_tokens_details": {
                    "reasoning_tokens": 554
                }
            }
        },
        "compress_result": [
            {
                "video_url": "https://las-ai-cn-beijing-baseline.tos-cn-beijing.volces.com/operator_cards_serving/public/qa/las_vlm_video/v1/eating_56.mp4",
                "compression_details": {
                    "original_info": {
                        "original_path": "https://las-ai-cn-beijing-baseline.tos-cn-beijing.volces.com/operator_cards_serving/public/qa/las_vlm_video/v1/eating_56.mp4",
                        "original_size_mb": 0.35,
                        "original_width": 640,
                        "original_height": 360,
                        "original_fps": 25,
                        "original_duration": 2.9,
                        "original_format": "mp4"
                    },
                    "compression_process": {
                        "step_0_format_conversion": {
                            "executed": false
                        },
                        "step_1_fps_reduction": {
                            "executed": false
                        },
                        "step_2_resolution_reduction": {
                            "executed": false
                        },
                        "step_3_bitrate_compression": {
                            "executed": false
                        }
                    },
                    "final_result": {
                        "final_width": 640,
                        "final_height": 360,
                        "final_fps": 25,
                        "final_duration": 2.9,
                        "final_size_mb": 0.35,
                        "final_format": "mp4"
                    },
                    "timing_info": {
                        "total_processing_time_seconds": 0.23
                    }
                }
            }
        ]
    }
}

Error codes

HttpCode

Error code

Error message

Description

401

Authorization.Missing

Missing Authorization.

Authentication missing

401

ApiKey.Invalid

The api key is invalid.

API key is invalid

400

Url.Invalid

The url is invalid.

Video URL is invalid

Last updated: 2026.06.04 10:54:03