MirrorProof
Start

Methodology · current engine

Useful evidence, honest limits.

MirrorProof measures visible consistency between a virtual clothing preview and a distinct physical-wearing photo. It does not measure physical fit.

How the comparison works

The engine first confirms the images are distinct and usable. Only then does it calculate a score.

  1. Validate image signature, dimensions, lighting, and sharpness
  2. Reject a downloaded preview reused as the reality photo
  3. Correct orientation and normalize both images
  4. Compare central garment colour and edge profiles
  5. Calculate five weighted visible-consistency metrics
  6. Generate deterministic explanations and confidence

Versioned score formula

The weights live in one configuration file and are stored with the engine version for each result.

30%

Colour consistency

Normalized colour distributions in the central garment region.

25%

Pattern and texture

Garment-region edge density and visible structure.

20%

Silhouette consistency

Horizontal and vertical edge profiles after normalization.

15%

Garment length

Relative lower-torso coverage using row profiles.

10%

Person preservation

Visible upper identity-region consistency when available.

Confidence before score

Input conditions shape whether the system is confident enough to show a comparison at all.

Input qualityFraming, torso visibility, resolution, sharpness, lighting, profile alignment, and obstruction risk shape confidence.
Not physical fitThe score cannot determine comfort, exact sizing, fabric quality, stretch, construction, medical suitability, or movement.

Transient comparison

The live comparison route processes the submitted comparison images in server memory and does not upload them to Supabase.

YouCam separately processes the source photo and garment image to generate the virtual preview.
90–100
Highly reliable
75–89
Mostly reliable
60–74
Noticeable differences
0–59
Significant visual mismatch
Start a verification