Struct ImageAnalysisTask 

pub struct ImageAnalysisTask { /* private fields */ }

The scene-analysis task. Construct via ImageAnalysisTask::new.

Implementations

impl ImageAnalysisTask

pub fn new() -> Self

Construct with accept_empty = false (a payload that lacks the required indexable content — description AND tags both populated, OR at least one of the substantive detection buckets subjects / objects / actions non-empty — is treated as a model regression and rejected; see Self::with_accept_empty for the full predicate and the opt-in alternative).

pub const fn accept_empty(&self) -> bool

Returns whether the parser accepts payloads that lack the required indexable content (description AND tags both non-empty). See Self::with_accept_empty for the trade-off.

pub const fn with_accept_empty(self, val: bool) -> Self

Builder-style setter for accept_empty.

When false (default), the parser rejects payloads that lack the required indexable content as JsonParseError::NoUsableFields. The composite threshold accepts a payload when either:

• description AND tags are both populated (the prose + keyword path; matches the integration-test smoke criterion), OR
• at least one of the substantive detection buckets — subjects, objects, or actions — is non-empty (the substantive-detection path; preserves who/what/where search metadata even when the model fails to summarize).

Style/attribute buckets (mood, lighting) and single-label fields (scene, shot_type) are intentionally NOT in the substantive path. A payload like lighting: ["natural light"] or mood: ["calm"] alone (description and tags empty, no substantive detections) is more often a regression than a legitimate weak-but-real scene; rejecting it surfaces the failure instead of writing a single-attribute stub to the search index.

Tags-only, scene-only, description-only, shot_type-only, mood/lighting-only, and fully-empty payloads all fail both paths and are rejected. This is the right setting for indexing pipelines: it surfaces decoder/model regressions that would otherwise silently overwrite real metadata with sparse search records.

When true, the parser bypasses the indexable-content check and returns whatever round-trips through the schema. IMAGE_ANALYSIS_PROMPT explicitly tells the model to “Use empty arrays or empty strings when a field is unknown”, so on truly low-information frames (blank, fade-to-black, plain color) compliant model output can legitimately be sparse or fully-empty. Use this knob if your pipeline distinguishes “low-information scene” from “no useful content” via something other than the parser (e.g. scenesdetect’s keyframe scoring).

pub const fn set_accept_empty(&mut self, val: bool) -> &mut Self

In-place setter for accept_empty. See Self::with_accept_empty for the trade-off.

Trait Implementations

impl Clone for ImageAnalysisTask

fn clone(&self) -> ImageAnalysisTask

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Default for ImageAnalysisTask

fn default() -> Self

Returns the “default value” for a type.

impl Task for ImageAnalysisTask

type Output = ImageAnalysis

The typed result of a successful run.

type Value = Value

The schema/grammar value the Task carries. Typically:

type ParseError = JsonParseError

The error type returned by Task::parse. JSON-parsing Tasks typically use crate::JsonParseError (behind the json feature).

fn prompt(&self) -> &str

The user-message prompt sent alongside the images.

fn schema(&self) -> &Value

Borrow the schema/grammar value. Zero-cost typed access for engines that bind Task::Value to a concrete type. Engines that need a unified Grammar enum should call Task::grammar instead.

fn grammar(&self) -> Grammar

Constrained-decoding grammar for this task, wrapped in the engine-agnostic Grammar enum.

fn parse(&self, raw: &str) -> Result<Self::Output, JsonParseError>

Parse the model’s raw text output into a typed Output.