원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
AI Tasks Skill
When to use
- User wants to tag images with business-specific categories
- User needs controlled vocabularies (predefined value lists)
- User wants yes/no quality checks on images
- User needs to extract metadata using natural language analysis
AI Task Structure
Each AI Task contains 1-10 sub-tasks with:
- Instruction (required): Natural language question about the image
- Action Type (required):
select_tags,select_metadata, oryes_no - Vocabulary (optional): 1-30 predefined approved values
Action Types
select_tags
Adds tags from vocabulary. Supports multiple selections.
{
"type": "select_tags",
"instruction": "What body style is this vehicle?",
"vocabulary": ["sedan", "suv", "hatchback"],
"max_selections": 1
}select_metadata
Sets custom metadata field (field must exist in DAM).
{
"type": "select_metadata",
"instruction": "What is the primary color?",
"field": "primary_color",
"vocabulary": ["red", "blue", "white", "black"],
"max_selections": 1
}yes_no
Binary quality check with conditional actions.
{
"type": "yes_no",
"instruction": "Is the product completely visible?",
"on_yes": { "add_tags": ["framing_ok"] },
"on_no": { "add_tags": ["needs_reshot"] }
}Gotchas
- Instruction clarity: Be specific ("What is the collar type?" not "Describe the image")
- Vocabulary design: Use business terminology, non-overlapping, 1-30 items max
- Field requirements: For
select_metadata, field must exist in DAM first - Tag values: Cannot contain
%character - yes_no tasks: Must have at least one of
on_yesoron_no - Vocabulary length: Max 500 characters combined (select_tags only)
- Scope: Works on visual content only (images/videos)
Applying AI Tasks
- Via Saved Extensions: Create and apply via dashboard/API
- Via API at Upload: Include in
extensionsarray - Via Path Policies: Auto-apply to files in specific folders
Full examples
For complete, copy-ready ai-tasks configurations organized by industry (fashion e-commerce, travel, automotive) and detailed per-task-type parameter references, read resources/EXAMPLES.md.

