All Products
Search

This Product

    Platform For AI:Multimedia analysis

    Document Center

    Platform For AI:Multimedia analysis

    Last Updated:Jun 23, 2026

    Multimedia analysis provides out-of-the-box algorithm capabilities for images and videos, including foundation model services and advanced model services.

    Background information

    Multimedia analysis supports the following algorithm services:

    • Foundation model services: Out-of-the-box algorithm capabilities for images, including multi-label image tagging, image quality assessment, facial attribute analysis (such as attractiveness, face shape, hairstyle, and hair color), age analysis, figure modification (slimming or plus-size), and watermark removal.

    • Advanced model services: Out-of-the-box algorithm capabilities for videos, including video classification and tagging, video quality assessment, dynamic classification and tagging for posts with images and videos (used for tagging multimodal content such as dynamic posts and threads), and AI-generated image tagging. The tags improve the training of AI image generation models.

    Billing details

    Multimedia analysis supports two billing methods: pay-as-you-go and subscription resource plans. For more information, see Billing details for multimedia analysis.

    Usage guide

    Activate multimedia analysis and purchase a resource plan

    First-time users must activate the service in the Multimedia Analysis section under Solutions on the Platform for AI (PAI) page.

    1. Log on to the PAI console.

    2. Follow the instructions in the figure to activate the Multimedia Analysis service.

    3. The pay-as-you-go billing method is used by default. You are billed based on the number of calls.38e9535689e0b041a5c5c5a0ca32dd1a.png

    You can also purchase a resource plan with a one-time payment for a lower price.

    1. On the Basic Model Service tab of the Multimedia Analytics page, click Purchase Resource Plan.

    2. On the Subscription Model Service page, configure the Quantity, Scenarios, API Calls and Duration parameters, and then click Buy Now.

    3. To use multimedia analysis services, set the Scenarios parameter to Multimedia Analysis-Basic Model Service or Multimedia Analysis-Advanced Model Service. Configure the other parameters based on your business requirements.

    Python SDK instructions

    After activating the multimedia analysis service, you can use the Python SDK to call various algorithm services. For more information, see Multimedia analysis: Python SDK instructions.

    Java SDK instructions

    After activating the multimedia analysis service, refer to the Java SDK GitHub for details about using the Java SDK to call algorithm service API operations. The Java SDK parameters are nearly identical to those of the Python SDK. For parameter details, see Multimedia analysis: Python SDK instructions.

    Multimedia analysis capabilities matrix

    Specification

    Model service name

    Consumption per service call

    Description

    Example

    Foundation model service

    Image quality assessment

    1 foundation model service call

    Provides image quality assessment and returns a floating-point score from 0 to 100.

    "iqa_result":66.88

    Facial attribute analysis

    1 foundation model service call

    • Outputs facial attributes, including face shape, hair color, hairstyle, and attractiveness.

    • Differentiates multiple faces by the coordinates of facial regions. If no face is detected, an empty array is returned.

    • Face shape: Triangle, Round, Heart, Square, Oval, Diamond, Long.

    • Female hairstyle:

      • Bangs type: Center-parted, Braided, Side-swept, No bangs, Wispy, Blunt.

      • Curl type: Cloud curls, Large waves, Small waves, Airy curls, Permed curls, Frizzy curls, Egg-roll curls.

      • Hairstyle: Curly, Updo, Straight, Ponytail, Braided.

      • Hair length: Medium, Short, Long.

    • Male hairstyle: Parted, Buzz cut, Crew cut, Flat top, Butch cut, Textured crop, Layered, Slicked back.

    • Hair color: Black, Coffee, Ash gray, Chestnut, Brown, Gradient, Burgundy, Gold, Yellow, Other.

    • Attractiveness: 0 to 5.

    Age analysis

    1 foundation model service call

    • Detects the age range of the main face in an image.

    • If there are multiple faces in the image, only the result for the largest face is returned. If no face is detected, an error is returned.

    Age ranges include the following: '0-2', '3-9', '10-19', '20-29', '30-39', '40-49', '50-59', '60-69', and '70+'.

    Multi-label image tagging

    1 foundation model service call

    Provides multi-label image tagging. Outputs the top K tags with the highest probabilities and their corresponding high-dimensional features.

    Examples of frequent tags: girl, selfie, boy, daily life, screenshot, food, car, cuisine, game, cartoon, animal, Korean fashion.

    Figure modification

    1 foundation model service call

    Modifies the figure in a portrait. Upload a portrait and adjust the figure by changing the degree parameter to make the figure slimmer or larger. A degree > 0 indicates slimming.

    Returns the Base64-encoded modified image.

    Watermark removal

    1 foundation model service call

    Removes watermarks from an image.

    Returns the Base64-encoded image after watermark removal.

    AI-generated image tagging

    1 foundation model service call

    Provides multi-label image tagging capabilities for training AI image generation models, such as Stable Diffusion. Better tags improve the quality of the generated images.

    • Supported tagging models: WD14, BLIP, GIT, RAM.

    • Example caption result:

      "sensitive, 1girl, solo, long hair, looking at viewer, smile, black hair, brown eyes, scarf, lips, realistic".

    Custom model service

    N foundation model service calls. The value of N varies based on the complexity of the custom model.

    Provides custom model services for images and videos.

    Depends on the specific type of custom model.

    Advanced model service

    Dynamic classification and tagging for posts with images and videos

    1 advanced model service call

    Classifies and tags dynamic posts or threads containing multimodal content. Supports text-image and text-video combinations. Also returns high-dimensional feature embeddings.

    • Examples of frequent classes: life, movies and TV shows, sports, travel, games, food, fitness.

    • Examples of frequent tags: sports, food, dance, fitness, cooking, travel, selfie.

    • Example embedding:

      0.915,0.882,0.943,0.978,1.027,1.181,1.066,1.029,0.866,0.716,0.628,1.203,0.689,0.533,0.734,1.038,0.98,0.613,0.96,0.88,0.586,0.702,1.515,0.697,0.987,0.699,1.179,4.274,0.757,0.89,0.805,0.901.

    Video quality assessment

    1 advanced model service call

    Assesses short video quality and returns a floating-point score from 0 to 100.

    "video_score":20.57

    Video classification and tagging

    1 advanced model service call

    Classifies and tags short videos. Returns the video class and the top K tags with the highest probabilities. Also supports outputting high-dimensional video features.

    • Examples of frequent classes: life, knowledge, music, technology, games.

    • Examples of frequent tags: with captions, girl, social news, slimming and shaping, skits, movie clips, natural scenery.

    Testing and service

    For testing and support, contact us by submitting a ticket.