Jin Engine v2.0 is Live

Predictable 360° virtual try-ons. Zero camera crew.

Jin uses agentic AI and 3D Gaussian Splatting to turn flat-lay garments into cinematic, physics-accurate ramp walks for enterprise retail.

Render · 360° · 24fps

The pipeline

Deterministic in. Cinematic out.

Three agentic stages, each one validated before the next runs.

ingest · garment.schema.json
1{
2 "sku": "JKT-4471-BLK",
3 "garment": "structured_jacket",
4 "fabric": {
5 "composition": "wool_blend",
6 "weight_gsm": 412,
7 "drape_coeff": 0.31,
8 "stretch": 0.04
9 },
10 "seams": 14,
11 "confidence": 0.987,
12 "validated": true
13}

Deterministic data ingestion.

Every flat-lay is parsed into a Zod-validated garment schema — fabric weight, drape coefficient, seam map — before a single frame is rendered.

splat · 1.2M pts

Spatial 3D mapping.

Gaussian Splatting reconstructs the garment as a true volumetric model, not a 2D warp. Every angle is a real angle.

qa-agent · live
jin-qa — 80×24

Self-correcting QA.

An agentic review loop checks physics, seam alignment and temporal consistency, then re-renders anything below threshold — automatically.

Before / After

From static asset to spatial video in 45 seconds.

Drag to compare the ghost-mannequin source with the Jin render.

Ghost-mannequin product photo of a jacket
Jin-rendered spatial video frame of the same jacket
Source · ghost mannequin
Jin · frame 128 / 360