Geometry you can question.

How a reconstruction works

Your browser checks each photograph and runs the locally served MediaPipe Face Landmarker. Multiple angles contribute to the facial analysis: the server aligns and robustly fuses their landmarks, then calculates documented measurements.

When complete-bust reconstruction is available, you choose one appearance photograph per person. The browser removes its background and prepares a head-and-shoulders image. After your consent and Reconstruct action, Somaiva temporarily processes that prepared image on Google Cloud to generate a textured bust, including ears, neck, shoulders and hair geometry.

The selected photograph determines the hairstyle and clothing. Other photographs inform the separate facial measurements. Unseen anatomy, the back of the head and hidden hair are inferred; plausible completion is not a verified likeness of those regions. Color and Mesh display the same bust geometry. Landmark inspection, overlay, difference map and morph use the independent facial analysis mesh.

Earlier sessions and the facial-analysis-only workflow retain their fitted face and generic cranium. They can be reconstructed again when the complete-bust engine is available.

What the numbers mean

Measurements describe external soft-tissue and surface geometry. Scale is normalized so facial height equals 1. There is no verified metric calibration, and values are not millimetres.

Depth is partly inferred by the landmark model. Unknown lenses, camera distance, makeup, retouching, age differences, expression and occlusion can change the result. Coverage scores and confidence proxies do not establish surgical accuracy.

MediaPipe supplies no calibrated confidence for every landmark. The displayed proxy combines conservative quality and consistency signals. Unsupported measurements are withheld. A realistic-looking bust does not improve or validate these measurements.

What it cannot tell you

Photographs do not reveal skull structure. This tool cannot determine which procedures you need, predict clinical outcomes, identify a person, or establish attractiveness or gender.

Gender-perception model not yet available. No masculinity, femininity or “passing” rules are hidden in these measurements.

The 3D editor offers local shape controls, reference-nose blending and optional aesthetic presets, including a feminine-inspired preset. These are adjustable styling choices, not gender classification, medical recommendations or surgical outcome predictions. Edits are saved as separate simulations; original measurements remain unchanged.

What is processed and saved

Your original files and the other photographs stay on your device. For a complete bust, only the selected background-removed, metadata-free image is uploaded, together with facial landmarks needed for registration. Processing uses Somaiva's infrastructure on Google Cloud, without another image-generation service. Selecting files alone does not send images or landmarks.

The prepared image and registration landmarks are scheduled for deletion as soon as processing finishes, fails or is cancelled. After your browser saves the downloaded bust, it requests deletion of the server result. Failed cleanup can be retried; abandoned jobs become inaccessible after 24 hours and are cleaned up on a best-effort basis. A temporary access token in this tab allows resuming after reload. It stores no photographs.

Meshes, textures, landmarks, measurements, names and dates are saved in this browser's IndexedDB. Delete Session removes that local record; clearing site data removes all local sessions. GLB downloads include the complete textured bust; JSON contains the separate facial geometry. Downloaded files are under your control. Original previews are released after saving or leaving the upload page.

The facial-analysis-only workflow sends landmarks for processing in memory and assembles its optional color texture locally. Local text commands and shape editing stay in your browser. If the optional AI text planner is enabled, choosing it and consenting sends your command and slider values to Google Cloud; it does not send photographs, geometry or session names. Neither workflow trains on your uploads or uses analytics. Application logs exclude image bytes, landmark arrays and edit prompts.

Suitable photographs

Use 3–60 JPEG, PNG, WebP or AVIF photographs of the same consenting adult. Include frontal, left and right three-quarter views; profiles can improve facial coverage when detectable. Use a neutral expression and unobstructed features.

For the bust appearance reference, choose a clear front or slight three-quarter photograph that includes the entire hair or scalp, ears, neck and shoulders. If hairstyles differ, choose the photograph whose appearance you want to reconstruct; the generator does not average those hairstyles.

HEIC is not supported in this version; export it as JPEG. Each image is limited to 20 MB and 32 megapixels and is resized locally. Duplicate photographs are excluded. Repeated similar poses add little coverage; additional images do not guarantee greater accuracy. Optional guided capture checks seven distinct usable angles across both sides, but does not calibrate physical scale or earn a higher confidence score.

Built for evidence

Each measurement exposes its landmark definition in the geometry lab. Comparison uses stable-landmark alignment and optional global scale normalization; it does not force entire surfaces into agreement. The generated bust is registered to that facial coordinate system for viewing.

Synthetic benchmarks test deterministic geometry and regression, but do not prove identity fidelity or clinical accuracy. In the procedural-head image benchmark, MediaPipe detected all five rendered views, yet chin projection error reached 78%, nose projection 29% and jaw width 11%. These results describe the facial analysis engine, not the generated bust. Real multi-view ground truth and repeat captures are needed to establish anatomical accuracy.

Reconstruction, appearance analysis and future human review are separate interfaces. Payments or referral relationships cannot alter geometric measurements.