The function of the MPS Demo is relatively simple, only for experiencing the basic effect, please use the API access to test the complete effect.
1. Open MPS.LIVE, enter the LLM Summarize experience page, select Offline Video (Offline File) or Live Streaming, and click One-Click Processing.
2. Once the processing is complete, you can view the results.
API Integration
Initiating a Summary Task
Call the Media Processing Service (MPS) API, select AiAnalysisTask, set Definition to 22 (preset large language model (LLM) summarize template), and enter extended parameters in ExtendedParameter for specific capabilities. For details, see Extended Parameter Description below.
Example:
{
"InputInfo":{
"Type":"URL",
"UrlInputInfo":{
"Url":"https://facedetectioncos-1251132611.cos.ap-guangzhou.myqcloud.com/video/xxx.mp4"// Replace it with the URL of the video to be summarized.
You can perform quick verification through API Explorer. After filling in relevant parameter information on the page, you can initiate an online API call.
Extended Parameter Description
ExtendedParameter is used to personalize the summary task, which can be left unfilled, combined with the default effect, and used on-demand for directions that need to be improved.
Note:
API Explorer will automatically convert the format. You only need to enter the corresponding ExtendedParameter in JSON format without converting it to a string. If calling the API directly, you need to escape the JSON string.
For the complete list of ExtendedParameter's optional parameters and their descriptions, refer to the following table:
{
"des":{
"split":{
"method":"llm",
"model":"deepseek-v3",
"max_split_time_sec":100,
"extend_prompt":"This video is a medical scenario video, which is segmented according to domain-specific medical knowledge points."
},
"need_ocr":true,
"ocr_type":"ppt",
"only_segment":0,
"text_requirement":"summary is within 40 characters",
"dstlang":"zh"
}
}
Parameter
Required
Type
Description
split.method
No
string
Segmentation Method: llm indicates Large Language Model-based segmentation, nlp indicates traditional NLP-based segmentation. The default value is
split.model
No
string
Segmentation llm: Available options include Hunyuan, DeepSeek-V3, DeepSeek-R1. The default value is
split.max_split_time_sec
No
int
Forces the maximum segmentation time in seconds to be specified. It is recommended to use it only if necessary, it may affect the segmentation effect. The default value is 3600.
split.extend_prompt
No
string
Requirements for segmentation task prompts. For example: "This instructional video is segmented by knowledge points". It is recommended to initially leave blank for testing and supplement prompts only when results fall short of expectations.
need_ocr
No
bool
Whether to use Optical Character Recognition (OCR) to assist segmentation,
If disabled, the system only recognizes the video's speech content to assist in video segmentation; if enabled, it also recognizes the text content on the video image to assist in video segmentation.
ocr_type
No
string
OCR auxiliary type：ppt: Processes on-screen content as PowerPoint slides and segments videos based on slide transitions. other: Applies alternative segmentation methods.The default value is ppt.
only_segment
No
int
Whether to only segment without generating a summary. The default value is 0.
1: Only segment without generating a summary.
0: Segment and generate a summary.
text_requirement
No
string
Requirements for generating a summary. For example, the character limit is "summary is within 40 characters".
dstlang
No
string
Title and summary language. The default value is "zh".
"zh": Chinese.
"en": English.
Querying Task Results
Task callbacks: When initiating an MPS task using ProcessMedia, you can set callback information through the TaskNotifyConfig parameter. After the task is completed, the task results will be called back through the configured callback information. You can parse the event notification results through ParseNotification.
Description corresponds to the entire video summary, and Paragraphs corresponds to the intelligent segmentation results of the entire video and the summary of each segment.
