NOSEJOB.APP
Low-key black-and-white profile lit from the left along the nose and cheek

Technology

Precision,
without pretence.

An engineering approach to aesthetic simulation: local analysis, region-locked generation and compositing that protects everything you didn't ask to change.

The pipeline

From portrait to possibility.

  1. 01

    On-device landmarking

    MediaPipe Face Landmarker runs in your browser via WebAssembly and locates 478 points, including iris centres. Your photo is not uploaded for analysis.

  2. 02

    Quality & pose checks

    Head yaw, pitch and roll are estimated from the landmark geometry. Portraits that are too rotated, too small or show several faces are flagged before any AI is used.

  3. 03

    Region masks

    Each agent owns anatomical regions. Landmarks become a feathered, dilated mask — nose, lips, chin, jawline, cheeks, periorbital area, skin, teeth or hairline.

  4. 04

    Face-centred crop

    Only a square crop around the face is sent for editing, at up to 1024 px, re-encoded to strip camera metadata such as GPS location.

  5. 05

    Provider-agnostic generation

    The server builds the instruction from trusted agent definitions and calls the configured image model: OpenAI, FLUX Fill via Replicate or fal, or Gemini.

  6. 06

    Identity-preserving composite

    The generated pixels are blended back onto your full-resolution original through the feathered mask. Outside the mask, the image is unchanged pixel for pixel.

2D analysis vs 3D reconstruction

We tell you exactly
what we measured.

A single ordinary photograph contains two-dimensional information. Our analysis measures landmark positions and ratios in that image plane — that is 2D facial landmark analysis.

The landmark model also estimates a relative depth value for each point. We visualise it as an estimated depth mesh, explicitly labelled as such. It is not metric, not a scan, and not a substitute for clinical 3D imaging.

Adding left and right side-profile photographs gives the studio additional views to analyse and edit — but we never describe the result as a true 3D face scan.

What we do

  • 478-point 2D landmarking
  • Proportion & symmetry ratios
  • Head pose estimation
  • Estimated relative depth mesh
  • Multi-view (front + profiles)

What we don't claim

  • A true 3D face scan
  • Millimetre measurements
  • Surgical predictions
  • Guaranteed outcomes
  • Medical diagnosis
Monochrome side-profile close-up highlighting the jawline and an ear hoop

Experience it on your own portrait.