Pipeline Calibration

Deterministic pre-flight benchmark evaluating storage, schema, compiler, models, and WebCodecs.

The Pipeline Calibration engine (services/calibration/pipelineCalibration.ts) in Serelius Flow Video provides end-to-end deterministic diagnostics before running generation batches. It validates browser runtime subsystems, prompt compilation, local database integrity, and Google Flow API constraints to eliminate runtime failures and protect workspace credits.


1. Interface Integration

Calibration is directly accessible from the main interface and through automated pre-flight checks:

  • Header Calibration Button: Located in the top bar adjacent to SETTINGS. Clicking CALIBRATION opens the ENTERPRISE PIPELINE CALIBRATION modal, running an immediate diagnostic benchmark or allowing manual triggers via EXECUTE FULL CALIBRATION.
  • Automated Pre-Flight Toggle: Configured under SETTINGS > AUTOMATION via PRE-FLIGHT CALIBRATION (YES / NO). When enabled, clicking START executes the benchmark before launching automated batches, surfacing warnings or blocking errors with a direct VIEW CALIBRATION shortcut.

2. The 5 Deterministic Verification Pillars

The diagnostic engine evaluates five operational layers without consuming AI generation credits:

PillarSubsystemVerification ScopeTarget Checks & Fallbacks
StorageIndexedDBSERELIFLOW_VIDEOS_DBVerifies active database connection and transactional integrity across project_metadata, media_blobs, and app_meta stores. Emits a warning (WARN) if sandboxing forces a fallback to localStorage.
SchemaScript StructureScene Matrix & CharactersValidates canonical character registry, sequential scene indexing, and flags previously failed scenes (ERRO) requiring manual creator review.
CompilerPrompt EngineReal-Time Test CompilationCompiles sample test scenes validating framing tags (CENTER, LEFT, RIGHT), aspect ratios (16:9, 9:16), visual styles, and acoustic consistency locks (ACOUSTIC & VOICE CONSISTENCY LOCK).
ModelsFlow CatalogModel Compatibility & ResolutionValidates active image models (Nano Banana Pro, Nano Banana 2.1, Nano Banana 2 Lite) and video models (Omni 1.1 Flash, Veo 3.1). Enforces 720p constraints for Veo 3.1 with auto-adjust warnings (WARN).
VideoMedia ProcessingWebCodecs & Hardware AccelerationInitializes @mediabunny/ac3 and @mediabunny/mp3-encoder, confirming hardware acceleration readiness for burned subtitles, optical transitions, and audio remuxing.
interface CalibrationResult {
  status: "PASS" | "WARN" | "FAIL";
  totalLatencyMs: number;
  pillars: {
    storage: { status: string; latencyMs: number; message?: string };
    schema: { status: string; latencyMs: number; message?: string };
    compiler: { status: string; latencyMs: number; message?: string };
    models: { status: string; latencyMs: number; message?: string };
    video: { status: string; latencyMs: number; message?: string };
  };
}

3. Calibration Results and Severities

The calibration suite records individual subsystem latencies and returns an overall health status in tens of milliseconds:

StatusColor BadgeOperational MeaningAction Required
PASSGreenAll five subsystems are 100% operational.Ready for automatic or manual generation.
WARNAmberPipeline is functional with non-blocking constraints or fallbacks active.Review notices (such as enforced 720p resolution or localStorage mode).
FAILRedCritical blocker detected in schemas, models, or runtime environment.Correct structural errors or missing models before spending credits.
Nuxfire Production Kit

Ready to build and launch your SaaS?

Get 100% full source code ownership, no proprietary wrappers, with architecture ready for millions of requests on Cloudflare.

© 2026 Nuxfire