Sharp Stock

English

简体中文

English

简体中文

AI Smart Description Generation ​

Sharp Stock deeply integrates advanced multimodal large language models to automatically transform the content of image or audio assets into searchable, high-quality text descriptions.

1. Core Logic Description ​

When you click the "Auto" button with the blue AI icon in the detail panel, the system executes the following process:

  1. Source File Reading: The system directly reads the original asset file on the disk.
  2. Model Call: Depending on the asset type, the system automatically calls the interface defined in [Settings] → [AI Model Configuration]. Image assets are sent to the vision model, while audio assets are sent to the audio recognition model.
  3. Content Generation: The raw description returned by the model is optimized by the system and automatically filled into the note text box.

NOTE

The current version primarily supports standalone file assets. AI generation for Spine or Sprites aggregated assets must be applied to their associated underlying source files.

2. Operation Guide ​

  1. Interface Configuration: Go to [Settings] → [AI Model Configuration] and fill in the API address, model name, and access token (API Key). Local Ollama services are supported to ensure intranet data privacy.
  2. Trigger Analysis: In the right detail panel, expand the "Notes" section and click the "Auto" button.
  3. Result Persistence: Upon successful generation, the description info is automatically written to the database. If you perform secondary manual edits on top of the generation, be sure to click the "Save" button.

Limitations:

  • Format Support: AI description entry is currently not available for PSD, ICO, and Sprites virtual assets.
  • Batch Processing: The current version only supports AI generation for a single selected asset and does not support batch queue processing after multi-selection.

3. Typical Use Cases ​

  • Rapid Asset Documentation: Quickly complete readable descriptions for scene screenshots, character illustrations, or ambient sound effects, making it easier for team members to understand.
  • Natural Language Retrieval: AI-generated descriptions are fully included in the global search index. You can directly input natural language keywords like "bright blue sky" to locate assets, even if the filename does not contain such information.

4. Performance & Privacy Strategy ​

  • Local Privacy Protection: By connecting to locally deployed models (e.g., Ollama), you can ensure that asset materials are not transmitted over the public internet, achieving the highest level of privacy protection.
  • Traffic Optimization: On the AI Model Configuration page, you can customize the maximum resolution of images uploaded to the AI interface to balance recognition accuracy and network traffic consumption.

NOTE

The quality of generated descriptions depends primarily on the capabilities of the selected model. Custom prompts (System Prompt) are not supported in the current version.