visionTokensPerImage property
How many tokens the engine encodes each image into, when that is a setting of the engine (EmbeddingGemma 2: 140 or 70). The same image gives another vector under another budget, so whatever stores image vectors keys them by it. Null when the engine has no such setting or serves no images; when set, positive and only with EmbeddingModality.image in modalities.
Implementation
final int? visionTokensPerImage;