ImageTiling class

How to split each attached image into tiles before prompt rendering.

Set on ModelSpec.imageTiling (or directly on LlamaChatClient) to have every image-typed DataContent replaced, in place, by columns x rows overlapping PNG crops. Each crop consumes its own vision-encoder token budget and its own media marker, so prefill cost scales with the tile count.

Constructors

ImageTiling({int columns = 2, int rows = 2, double overlap = 0.1, int minimumEdge = 512, bool includeFullImage = false})

Creates a tiling configuration; the default splits into 2x2 tiles with 10% overlap.

const

Properties

columns → int

Number of tile columns.

final

hashCode → int

The hash code for this object.

no setter, override

includeFullImage → bool

Whether to keep the untiled image ahead of its tiles, giving the model a whole-page view for layout plus the crops for detail. Costs one extra image budget per attachment.

final

minimumEdge → int

Images whose width AND height are both below this skip tiling and pass through unchanged — small images gain nothing from being split.

final

overlap → double

Fraction of a tile's width/height duplicated into each neighbouring tile, so text straddling a cut line appears whole in at least one tile. 0.1 duplicates 10% on each interior edge.

final

rows → int

Number of tile rows.

final

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

Methods

applies(Image image) → bool

Whether image is large enough to be worth splitting.

noSuchMethod(Invocation invocation) → dynamic

Invoked when a nonexistent method or property is accessed.

inherited

toString() → String

A string representation of this object.

override

Operators

operator ==(Object other) → bool

The equality operator.

override