Property DefaultImageDetail
DefaultImageDetail
Gets or sets the default ImageDetail applied to every vision-capable pipeline created from this point onward.
public static ImageDetail DefaultImageDetail { get; set; }Property Value
- ImageDetail
Default is High.
Remarks
This is the pixel budget an image receives before vision encoding,
which is what decides how many vision tokens it costs. Every LM-Kit
pipeline that feeds images to a model reads this value when it is
constructed: conversations (MultiTurnConversation,
SingleTurnConversation, and the PdfChat,
RagChat, AgentExecutor wrappers over them), VLM OCR,
document-to-markdown conversion, and document RAG page snapshots.
Setting it once therefore moves the whole SDK, and a host that
exposes an image-detail setting binds it here instead of assigning
the level onto each object it creates.
Individual instances can still override it through their own
ImageDetail member, and a pipeline that has a task-specific
reason to pick a level (language detection reads pages at
Standard, a markdown page retried after a
weak result is re-read at Maximal) keeps
that level. Changing this property does not affect objects that
have already been created.