🪄 Facemorph Tool

v12.0.15 · expression-aware fallback · Lucia PNG first

How-to & sign-off guide Master Q Master A

⚠️ Quality standards (Paddy directive — every output must satisfy these)

Zero noise + background REMOVED — all matting via fal-ai/birefnet (no MediaPipe staircase, no background bleed). Hardwired in code.
Clean borders — 12px Photoshop dilation stroke (not blur-threshold halo). Hardwired.
Lucia-style drop shadow — alpha 0.65, blur 90, offset (12, 32). Hardwired.
Sources prepped via Face Prep Tool — 1200×1200 canvas, eyes at (500,480)/(700,480), min 800px input, face ≥25% of frame.
Production path is Lucia's hand-Photoshopped PNG — canvas fallback is DEBUG/DRAFT only until Paddy signs it off.
Final assets must look like Lucia's reference — if not, reject and re-prep.

⏳ Loading AI face model…

📚 Face Library

Pick pre-aligned faces from Face Prep Tool — the source audit still checks direction and expression before output.

Face A — high-res front-facing photo

🧑

Click or drop photo

Use Google · Tools · Size · Large

Face B — high-res front-facing photo

🧑

Click or drop photo

Use Google · Tools · Size · Large

🪄 Morph PNG (Lucia's Photoshop output) — recommended

Upload the hand-Photoshopped morph for production quality. Transparent BG ideal; auto-segmented otherwise. REQUIRED for production. Without this, the tool can only make an explicit DEBUG/DRAFT fallback.

🪄

Click or drop morph PNG

PNG with alpha = best · JPG/PNG with white BG also OK

Use canvas fallback as a DEBUG/DRAFT preview only. It is not a proper production morph; upload Lucia's PNG for proper output.

Load Face A and Face B. The tool rejects unusable sources, warns on expression differences, and helps flip opposite-facing pairs.

Slide label:

Detecting facial landmarks (eyes, nose, mouth, ears, jaw)…

Computing alignment transform (rotate · scale)…

Removing backgrounds (fal-ai/birefnet HQ matting; MediaPipe fallback only if API fails)…

Building morph (Lucia PNG untouched, or explicit debug union-hull fallback)…

Applying white border + drop shadow (per spec)…

Rendering morph asset + multi-version grid…

💾 Save to KBT Library

Uploads Q-slot + answer pair to Drive and creates a draft kbt_question row (qtype=19). Question is then playable in the host app.

Q-slot asset — morph image (replaces [q] in Face Morph Q template)

Morph (sticker, transparent bg)

A-slot face A — clean Face A stickered (left position, transparent)

Face A only, sticker'd, positioned left of centre — for animated reveal

A-slot face B — clean Face B stickered (right position, transparent)

Face B only, sticker'd, positioned right of centre — for animated reveal

A-slot asset — both side-by-side (PREVIEW only; not used in final .pptx if you have animations wired)

A + B stickers, side-by-side (transparent bg) — preview/fallback

Individual people files — stickered (border + shadow) AND clean (no border)

Face A — stickered

Face B — stickered

Multiple versions — 6 preset feature mixes (each 1920×1080, click for link or PNG)

Paddy sign-off comparison — master reference vs current output

Master Q slide

Current Q-slot asset

Master A slide

Current A-slot asset

Approval checklist

Source pair is frontal, intentionally expression-matched or warning-accepted, and flipped/aligned where needed Morph is one believable face: no double mouth, double eye, or double ear bleed White border and drop shadow match the master reference closely enough Final Q and A assets are acceptable for the Face Morph template