Docs

Torch Object Storage

SenseFlow overview

Preset templates

TOS SenseFlow offers a variety of preset templates. This topic describes the currently available preset templates.

tip

SenseFlow is currently only available in China (Beijing). Please contact your account manager for access.

Preset template list

Template name	Description
Intelligent Security Template	This template is designed for security scenarios, such as video surveillance. It offers default content understanding capabilities for videos and images, as well as a daily summary workflow. It is ideal for business scenarios in which devices consistently generate and upload unstructured data to TOS buckets. Examples include industrial parks, retail stores, and residential areas, which usually involve large data volumes, multiple file types, and uploads from various locations.
Media Asset Management Template	This template is designed for video processing scenarios. It provides default content understanding capabilities for documents, images, and videos, along with a workflow for removing subtitles. It is ideal for long-term storage of unstructured data, such as images and videos, in settings like short drama platforms, film and television production companies, and content distributors. Examples include everyday photos, videos, and screenshots. These datasets are typically large in volume and contain rich metadata that requires long-term retention.

Intelligent Security Template

Dimension	Method	Description
Semantic video indexing	Vector indexing (VLM)	Captures frames from video clips, uses vision language models (VLMs) to understand them, and generates semantic descriptions that are then vectorized.
Image indexing	Multimodal indexing	Uses VLMs to understand captured images, and determines the appropriate type of alert to send.

This workflow automatically performs multimodal understanding on objects uploaded to the bucket and creates indexes. It supports three types of content: documents, images, and videos.

This workflow uses the content summarization operator to automatically generate a daily summary report of the previous day's content.

Media Asset Management Template

Dimension	Method	Description
Semantic scene indexing	Vector indexing (VLM)	Uses VLMs to understand photos and videos, and vectorizes the content for semantic searches.
Face indexing	Facial vector indexing	Performs face detection, extraction of facial features, and clustering, and supports search by person.

This workflow automatically performs multimodal understanding on objects uploaded to the bucket and creates indexes. It supports three types of content: documents, images, and videos.

This workflow automatically removes subtitles from video files. To use this workflow, you must specify where the outputs should be stored.

Last updated: 2026.09.16 10:48:06