Nominal turn order: front, left-facing profile, rear, right-facing profile, front. At 090 degrees the nose points image left; at 270 degrees it points image right. Left and right describe image-facing direction, not measured anatomical sides.
Nominal 45-degree directions; all eight views are generated interpretations at 4K, not measurements. Target pitch 0° and roll 0°. All eight body angles are generated from one front photo; they are interpretations, not measurements.
Attach the face anchor and the nearest matching angle. A back view establishes the rear silhouette, not unseen facial details.
Keep the relaxed standing shopping pose. Show front, 45°, true side and rear at equal scale, fully head to toe.
Pose references below show the baseline outfit. Apply these views to each new outfit; they are not shopping try-on results.
Front · 0° Stand upright facing the camera. Level shoulders, relaxed arms slightly clear of the torso, feet naturally spaced and uncrossed.45° · 45° Keep the relaxed shopping stance. Rotate the whole body together; no waist twist or exaggerated hip tilt.Side · 90° True profile: head, shoulders, hips and feet face the same direction. Keep the near forearm clear enough to show waist, side seam, hem and footwear.Rear · 180° Face directly away, with shoulders and hips square. Gently move hair over one shoulder if needed to reveal the back neckline and garment drape.
Keep natural posture and body proportions. No crossed legs, exaggerated back arch, hip thrust, tiptoe pose, waist twist or pulled hem. Footwear contacts the floor naturally.
Hands, sleeves, hair and accessories must not hide the neckline, waist, side seams, hem or footwear needed for the fit check. Hold the bag beside the outer thigh in front/45° views; omit it from side/rear fit views if it blocks clothing.
Keep her original facial appearance, age, skin tone under comparable lighting, hair and body proportions. Do not slim the body, lengthen legs or copy the store model.
Use side/rear product photos when available. When missing, label unseen construction as estimated and keep it plain; never present invented closures, cutouts, pockets or decoration as verified product details.
Capture and review rules
Consult the current Model 005 page and manifest before every generation. Attach original face and body source anchors; her own photos win conflicts. Generated turns guide pose only until My-le approves them. Product photos supply clothing only.
Four labeled full-body views: front, 45°, true side and rear. Use a spacious 2 × 2 sheet; provide individual views when detail needs more space.
Use full head-to-toe framing with margins and equal body scale, camera distance, level mid-body camera, background and soft neutral light across all four views.
Maintain the same outfit, colors, fabric, seam placement, hem length and footwear height. Rotate the same outfit with the body; never mirror asymmetric garment features.
Keep natural fabric drape. A generated shopping preview does not establish actual sizing, fabric performance or measured fit.
Reject:
Missing, duplicated or mislabeled view; side is not a true profile; rear is not directly away.
Cropped head or footwear, unequal scale, obstructed fit, distorted anatomy or floating feet.
Changed face, skin tone or body proportions; different outfit or inconsistent neckline, seams, hem or boots.
Unseen garment construction presented without an estimate label.
Reusable outfit-check prompt
Create a photorealistic outfit-check sheet of the same woman, My-le. Consult https://www.jbnx.io/design/m/005 and its current manifest first. Attach her original face source and body source as identity authorities, then the nearest body-angle references and the product photos for clothing only. Keep her facial appearance, age, skin tone, hair and exact body proportions. Wardrobe: {outfit from product references}. Four equally scaled head-to-toe panels in a spacious 2 × 2 layout: Front 0°, relaxed 45°, true Side 90°, Rear 180°. Rotate head, shoulders, hips and feet together. Stand naturally with uncrossed feet, relaxed shoulders and arms slightly clear of the torso; no exaggerated hip tilt, back arch, tiptoe pose, waist twist or pulling the hem. Keep hands, hair and bag clear of fit details; omit the bag from side/rear if needed. Same camera distance, level mid-body camera, soft neutral light and background. Preserve garment color, texture, seams, asymmetry, hem length and footwear across views. Label missing side/rear product construction as estimated; do not invent decorative features. Reject missing angles, obstructed fit, cropped feet, anatomy errors, identity/body/skin-tone changes and outfit drift. AI try-on; actual fit depends on sizing.
Closer reference
Face & expression
Front, slight smile Generated · acceptedDownloadThree-quarter, facing left Generated · acceptedDownloadProfile, facing left Generated · acceptedDownloadProfile, facing right Generated · acceptedDownloadThree-quarter, facing right Generated · acceptedDownloadSmile Generated · acceptedDownloadEyes closed Generated · acceptedDownloadSpeaking Generated · acceptedDownload
The approved front face is the face authority. Generated turns and expressions are interpretations that must match it. A speaking still is not a lip-sync model.
What stays fixed
Identity
Preserve her face and body. Change the scene around them.
My-le’s fixed identity features
Feature
Keep consistent
Face
Keep the face anchor's exact face shape, width and cheek fullness; never thinner, longer or older than her photos.
Eyes
Dark eyes; retain shape, spacing and eyelids from the face anchor.
Brows
Natural dark brows; retain shape and placement.
Nose
Retain the face anchor's nose proportions.
Lips
Retain mouth width, lip shape and her smile lines.
Skin
Her skin tone as in her photos, with natural texture; no beauty filter.
Hair
Long, loose, wavy black hair with a center part.
Body
Slim, toned build as in the body anchor. Do not slim, enlarge or lengthen body features.
Reference outfit
Fitted white short-sleeve crew-neck tee, black high-waist wide-leg trousers, plain white low-top sneakers.
Current face anchor · v0.1.0
Likeness scores · Jev and face-ID
ArcFace (InsightFace buffalo_l) cosine similarity against her real photos (each real photo scored against the others for calibration) and Jev (TypeSafe System One, jev-1.14.0) same-person probability and 0-4 likeness score with confidence against her face anchor. Confidence is not proof; My-le's own judgment decides.
Likeness against My-le’s approved face anchor
Image
Jev same person
Jev likeness (0–4) · confidence
Face-ID
Real photo · front neutral (control)
82%
3.09 · 0.26
0.52
Real photo · Chinatown (control)
78%
2.90 · 0.34
0.60
Different person · Bethany (control)
3%
0.19 · 0.85
0.02
Gate 1 face (rejected: longer, older)
83%
3.19 · 0.31
0.50
Gate 2 face (rejected: different woman)
75%
2.67 · 0.34
0.06
Face front
85%
3.08 · 0.25
0.50
Face three-quarter left
87%
3.35 · 0.44
0.53
Smile
85%
3.28 · 0.41
0.57
Body front
90%
3.53 · 0.61
0.53
Hero anchor
81%
2.95 · 0.42
0.46
Face-ID is ArcFace cosine similarity, mean against nine of her real photos; her own photos score 0.36-0.60 against each other and a different person 0.02. Jev compares each image with her restored face anchor. Gate round 2 shows the limit: Jev still gave a different woman 75%, while face-ID caught it (0.055). Confidence is the spread of Jev's answers, not proof. My-le decides.
What this version establishes
Her own photos are the identity originals and win any conflict. The owner chose the face anchor (smiling headshot) and the body anchor (striped tee + skirt); both are 4x restorations with face creativity 0. Generated references become authoritative only after My-le reviews them.
Unspecified: height, exact body measurements, ethnicity, age, unseen distinguishing marks.
Age direction: as photographed (as photographed); never made younger or older. Skin color is described as rendered in the references. The reference lighting is not a calibrated color measurement.
Real person
Consent & use
My-le is a real person, used with consent. Consent attested by owner (Joey Nguyen) on 2026-11-03. Scope: JBNX Design reference model page and AI-generated stills and video of her likeness.
Owner attests that My-le agreed to a public model page at jbnx.io/design/m/005 and to AI-generated stills and video of her likeness.
Subject review: My-le reviews new references and any clip of her speaking before production use.
Privacy: Only My-le is identifiable in provider inputs. Her 17 source photos are not published; only her-only crops used as inputs are recorded by hash.
Use only within the stated scope. Do not present generated images as photographs of real events.
Keep face, hair and body fixed; change scene and camera around them.
Start every shot here
Still & motion prompts
Build and review the scene first frame with the identity prompt. For image-to-video, describe movement around that accepted frame.
Generate {shot} of the same woman. Image 1 (her restored face anchor) is the face authority: keep her exact face shape, width and cheek fullness, eyes, brows, nose, mouth and smile lines, skin tone and long wavy black hair with a center part. Image 2 (nearest body view) sets her build. Never make her face thinner, longer or older than her photos. Wardrobe: {wardrobe}. Scene: {scene}. Lighting: {lighting}. Same real person, photorealistic, natural skin texture. One person only, no text, no watermark.
Animate the attached accepted scene first frame. Action: {one_simple_action} Camera: locked, no camera movement. Duration: 5 seconds. @myle is this same woman: keep her existing face, age, skin tone, long wavy black hair with a center part, body proportions, white tee, black wide-leg trousers and white sneakers, lighting, scene and ground contact stable throughout. Use natural movement and continuous anatomy. Do not add scene changes, new people, text, captions, logos or wardrobe changes. No dialogue. Avoid flicker, face drift, warped frames, body drift, floating heels, extra fingers and sliding feet.
Avoid: identity changes, longer, narrower, thinner or older face, generic beauty-model face, changed skin tone, hair color or part changes, plastic skin, beauty filter, facial warping, frame-to-frame flicker, body slimming, bulking or lengthening, wardrobe drift in reference shots, added text, captions or logos.
1. Anchor
Attach the face anchor. Add the nearest full-body angle.
2. Check a still
Compare face, hair and body in the first frame before generating the full shot.
3. Review motion
Check turns, walking, speech and the final frame. Reject face, age or body drift.
Identity drift checklist
Visible face matches her face anchor and real photos (shape, width, cheeks, eyes, smile).
Never thinner, longer or older than her photos.
Hair keeps the long black waves and center part.
Body keeps the anchor's proportions; clothing does not reshape them.
Skin tone stays consistent under intended lighting.
My-le reviews new references and any clip of her speaking before use.
Identity stays stable through motion, blinks and the final frame.
Record the reference files, prompt, generator version, settings and seed when available. A matching seed alone does not guarantee identity. Regenerate drifting outputs from the anchors.
A stable baseline
Capture standards
Reference image capture standards
Setting
Reference standard
Camera
Level camera at mid-body height · natural full-body perspective without wide-angle distortion
Framing
Full body, crown through shoes, level camera and consistent body scale
Light
Soft, even studio light without dramatic shadows
Background
Seamless medium warm-grey studio backdrop and floor
Expression
Calm closed lips unless an expression is briefed · gaze follows head direction
Head
Pitch 0° · Roll 0°
Reference outfit
Fitted white short-sleeve crew-neck tee, black high-waist wide-leg trousers, plain white low-top sneakers.
Pose
Neutral relaxed stance, arms slightly away from the sides, feet flat on one ground plane
Texture
Natural skin and matte textile detail; no beauty filter
Test before finishing
Production
Brief. Reviewed first frame. Short native-resolution test. QA. Finish only when needed.
Applied: AI Media Cost Framework · v3. 33 jobs quoted at $9.91; actual bills and setup/compute costs remain unknown.
Not setHeight · Real person; no height supplied and none is inferred from images.
Voice pendingowner will add it later
$10.00External API cap · no top-ups
Scene first frames
Walk / turn working frame · DownloadSit working frame · Download
Reviewed working inputs. Motion review: 4 accepted, 0 rejected. Motion baselines carry no audio.
Voice pending (owner will add it later)
Turn
accepted · 1/2 attempts
Lead-reviewed working motion baseline; shot-specific production review still required.
Deviation: Kling turned her toward image right and stopped at about three-quarter instead of a full left profile; accepted for identity and motion quality.
First frame: accepted working test input · QA: accepted
Walk
accepted · 2/2 attempts
Lead-reviewed working motion baseline; shot-specific production review still required.
Second attempt: she shifts back a step, then walks toward the camera; continuous but short travel. The first attempt jumped back mid-clip and was rejected.
First frame: accepted working test input · QA: accepted
Sit
accepted · 1/2 attempts
Lead-reviewed working motion baseline; shot-specific production review still required.
Talks to the camera with natural head motion and blinks; a hand briefly enters the bottom edge.
First frame: accepted working test input · QA: accepted
Voice pending (owner will add it later): no voice proof clip yet.
QA report · Complete for the stored v0.1.0 references. Future scenes require their own review. Next: My-le reviews the references and motion baselines; the owner adds her voice. Each production scene still needs its own first-frame and motion review.
Workflow, model & voice settings
References. Done in v0.1.0: restored face and body anchors, eight body angles, five face turns, three expressions, two scene frames and a hero anchor at 4K. My-le reviews before production use.
Shot brief. Define one action, scene, camera, lighting, wardrobe, aspect ratio, duration and the approved script, if any.
Scene first frame. Create one scene still. Review face, beard, glasses and anatomy before motion; record reviewer and image hash.
Motion test. Short native-resolution test from the accepted first frame. At most two attempts per shot.
Qa. Check face, beard, glasses, hands, hair, flicker, motion continuity and audio when present. Reject drift.
Final. Render 1080p only after an accepted test. Darren approves any clip of him speaking before release.
Hash the exact reference bytes, prompt, provider/model/version, settings, duration, resolution and audio options. Reuse accepted outputs; review existing outputs before retrying. A fixed seed is not an identity guarantee.
Cost per accepted deliverable includes all retries, discarded jobs, audio, LLM, finishing, storage and compute. Unknown cost stays unknown.
Renderer: kie · kling-3.0/video · kling-3.0 pro. Access: access verified actual jobs reviewed. Kling handles this real person's likeness; Seedance 2.x tops the Artificial Analysis image-to-video board but blocks real-person faces at the model layer. The 4K mode exists but adds cost and file weight for reference clips delivered at 1080p.
Stills: fal-ai/nano-banana-pro/edit at 4K. Restoration: fal-ai/topaz/upscale/image (Recovery V2 4x on the face anchor (creativity 0)).
Rates, resolution, sources, verification time and exact settings: production config.
Voice: ElevenLabs / · .
30- and 60-second finished video costs are not measured; quote from an approved shot list. Voice pending (owner will add it later).
Budget ledger
Observed consumption: 450 kie credits · $2.25 at listed rates. Actual cash charges remain unconfirmed.
Reserved/committed external API cap usage: $9.9100 of $10.00. This is not a final billed total.
face-000-g1 · rejected likeness gate · gate round 1: face read longer, narrower and older than her photos
body-000-g1 · rejected likeness gate · gate round 1: face read longer, narrower and older than her photos
anchor-g1 · rejected likeness gate · gate round 1: face read longer, narrower and older than her photos
face-000-g2 · rejected likeness gate · gate round 2: detailed face description produced a different woman (face-ID 0.055)
body-000-g2 · rejected likeness gate · gate round 2: detailed face description produced a different woman (face-ID 0.055)
anchor-g2 · rejected likeness gate · gate round 2: detailed face description produced a different woman (face-ID 0.055)
body-000 · accepted
body-045 · accepted
body-090 · accepted
body-135 · accepted
body-180 · accepted
body-225 · accepted
body-270 · accepted
body-315 · accepted
face-000 · accepted
face-045 · accepted
face-090 · accepted
face-270 · accepted
face-315 · accepted
expr-smile · accepted
expr-eyesclosed · accepted
expr-speaking · accepted
scene-floor · accepted
scene-chair · accepted
anchor · accepted
motion-turn · accepted
motion-walk · accepted
motion-sit · accepted
motion-talk · accepted
motion-walk-a1 · rejected continuity · attempt 1 rejected: she jumped back mid-clip (continuity break); 90 credits
jev-likeness · accepted
Full cost per accepted output remains unknown until cash billing and unmetered support costs are reconciled. Count video, audio, LLM, retries, discarded jobs, finishing, storage and compute.
Created from 17 photos the owner supplied, with My-le's consent attested by the owner. Face anchor: her smiling headshot; body anchor: striped tee + skirt (owner's choices), both restored 4x. Three 4K gate rounds (the first read longer and older than her photos; the second, with a detailed face description, became a different woman); round 3 accepted. 19 4K references, a 360 sheet and four Kling 3.0 pro motion baselines. Voice to come.
2026-11-03
Source anchors
Restored anchors the v0.1.0 stills were generated from. Kept for the record.