Mycelia Present · rendered 2026-05-25T15:27:45.484Z · source: ../convivium/clients/emerson_fry/EF_MODEL_PROFILE_AUDIT_v0.1.md

EF Model Profile Audit — v0.1

Trigger: Billy 2026-05-25 — supports the Model Profile system (UI spec: EF_MODEL_PROFILE_UI_SPEC_v0.1.md) that Faber is building so Remy can generate new editorial images using existing EF model imagery as reference. Purpose: identify EF's recurring models, characterize the photography register, propose the Profile data shape, flag likeness-rights, and recommend 2-3 starter profiles for the pitch deck. Method: systematic scan of emersonfry.com — homepage, blog index (~85 journal posts), main EF and Love Tòmas collections, individual product pages spanning multiple shoot eras. Image URLs extracted via JS + downloaded for visual inspection. Descriptors derived from observed visual content only — no speculation on real-world identities. Author: Mycelia (subagent → main session for review)

Executive summary

EF has run model-led editorial photography for ~10 years (earliest dated journal posts back to 2016). Over that period the brand has cycled through a small, recognizable cast — the founder/designer Emerson Fry herself appears as a "looking at it for u" mirror-selfie model throughout, and across that same span a rotating set of named/initialed models surface in editorial posts (Anna, Georgie/G, C, J, Elise T). In the current era (≈2023→present) two contemporary recurring models dominate: one sandy-blonde-blue-eyed (heart-shaped face) and one dark-haired Mediterranean/mixed-heritage (sleek low bun). These two cover ~70% of all on-model product photography across both Main EF (Mode A) and Love Tòmas (Mode B), and span the full setting range (interior atelier · outdoor coastal · farm-stand · garden · whitewashed villa).

The photography register is remarkably consistent: soft warm natural light, never studio strobes; intimate interior settings (NH/Maine farmhouse: white wainscoting + arched bookshelves + herringbone wood floors + French market baskets) OR outdoor coastal/garden/villa contexts; both models are typically slim-built and styled with neutral tonal hair (low bun / loose center-part) and tan-leather slides; mood is candid-contemplative, never high-energy, almost always looking down or in-3/4-profile rather than direct front camera. The aesthetic is intentionally not high-fashion-editorial; it reads home-grown / personal / quiet-luxury.

For the pitch deck, the two highest-leverage starter profiles are the current dominant pair — "M1 (Blonde)" and "M2 (Brunette)" — plus one historic profile, "Georgie/G" (a Black model who appeared across 2017-2019 EF + Love Tòmas imagery and represents the brand's broader styling latitude). All three have 8-15+ identifiable reference images at high resolution. Critical caveat: real people in these photos have publicity rights. The Profile system is for EF's own internal generation of imagery of their own models — not for external distribution or LoRA training — and EF should establish a written likeness-use consent flow with each model before using generated imagery in broad marketing. Faber's earlier flag stands.

§1 Recurring models identified

Method note

Each model identified here meets the threshold of appearing in ≥3 distinct shoots across the EF site. Descriptors are physical/visual only, never speculating on real-world identity. Where the EF blog gave a first-name or initial (Anna, G, C, J, Elise T), I preserve that label as a link to the source labeling — it's not a guess at real identity, it's what EF themselves used to caption their own posts. Internal model IDs (M1, M2, M3, etc.) are abstract.

M1 — "The Blonde / Atelier + Coastal" (Current dominant, both modes)

M2 — "The Brunette / Sleek-Bun" (Current dominant, both modes)

M3 — "Anna" (Historic, 2017-2019 era, primarily Main EF)

M4 — "Georgie / G" (Historic, 2017-2018 era, mixed mode)

M5 — "C / Beautiful C" (Historic, ~2020-2021 era, primarily Main EF)

M6 — "J" (Historic, ~2018-2020 era, primarily Main EF)

M7 — "Elise T" (Historic, ~2019 era, primarily Main EF)

M8 — "Emerson herself" (Founder / 2016-2022 era, both modes, mirror-selfie style)

Other recurring people surfaced but not model-class

These all surface in EF's tight-knit-family voice but are not photographed-as-models.

§2 Photography characteristics

Lighting

Setting (two anchor environments)

  1. "The atelier" (interior, ~70% of contemporary shots):
    • White wainscoting, paneled doors, arched bookshelves, sconces
    • Herringbone medium-tone oak wood floors
    • Antique wood dressers, French settees (cream brocade), wicker chairs
    • Decorative objects: candles, pottery bowls, French market baskets, blue-painted dishes, dried flowers, small flower arrangements
    • Reads NH or coastal-Maine farmhouse interior; intentionally not "studio"
  2. "The world" (exterior, ~30%):
    • Grass field / wildflower meadow
    • Lake dock + water
    • Farm stand with vegetables/produce + green doors
    • Whitewashed Mediterranean villa (Spain/Mallorca-style — used for Love Tòmas Vacation specifically)
    • Outdoor café table on gravel with iron furniture

Color treatment

Composition

Mood

Resolution / quality

Anti-photography (for the generator's exclusion list)

§3 Recommended Model Profile data shape

Aligned with the existing UI spec (EF_MODEL_PROFILE_UI_SPEC_v0.1.md) and tuned based on this audit:

type ModelProfile = {
  // Identity
  id: string;                          // 'model-001'
  internalName: string;                // 'M1' — internal-only, never client-facing
  displayLabel: string;                // 'The Blonde / Atelier + Coastal' — for admin UI
  createdAt: timestamp;
  updatedAt: timestamp;

  // Visual descriptor (purely physical/contextual — never identity-speculative)
  internalDescriptor: string;
    // e.g. "young woman mid-20s, sandy-blonde shoulder-length hair often
    //  worn straight with side part or low loose bun, heart-shaped face,
    //  full lips, blue-grey eyes, fair-but-sun-touched complexion, slim
    //  petite-medium build"

  // Reference images
  primaryRefImageURL: string;          // single best face-clear hero ref
  secondaryRefImageURLs: string[];     // 4-11 additional refs (variety in pose/setting/lighting)
  totalRefImageCount: number;          // computed
  refImageQualityNotes?: string;       // optional: "limited face-clear hero shots; most refs are 3/4 environmental"

  // Mode + collection context
  modes: ('A' | 'B' | 'both')[];       // Mode A (Main EF) / Mode B (Love Tòmas) / both
  signatureCollections: string[];      // e.g. ['ss26-mix-match-cotton-lace', 'lt-vacation', 'lt-caftans']
  signatureSeasons?: string[];         // e.g. ['SS26', 'FW25']

  // Style tags (drives prompt construction)
  settingTags: string[];               // ['interior-atelier', 'outdoor-coastal', 'mediterranean-villa', 'garden', 'farm-stand']
  stylingTags: string[];               // ['tan-leather-slides', 'french-market-basket', 'sunglasses-square-black', 'low-bun', 'minimal-jewelry']
  moodTags: string[];                  // ['intimate-candid', 'contemplative', 'looking-down', 'side-3-quarter']
  excludedSettings?: string[];         // settings the model has NOT appeared in (informs generation guardrails)

  // Source provenance
  sourceURLs: string[];                // EF page URLs where ref images live (for audit trail)
  firstSeenDate?: string;              // earliest dated appearance
  lastSeenDate?: string;               // most recent dated appearance
  appearanceCount: number;             // approximate, manually maintained

  // Consent + likeness rights (CRITICAL — see §4)
  consentStatus: 'none' | 'verbal' | 'written' | 'self' | 'unknown';
    // 'none': no consent obtained — DO NOT use for external imagery
    // 'verbal': verbal OK from model — flag for written follow-up
    // 'written': signed likeness-use agreement on file
    // 'self': model is the founder/team principal (self-consenting)
    // 'unknown': legacy model from older era; status unverified
  consentNotes?: string;               // free text: agreement scope, expiry, restrictions
  consentDocRef?: string;              // pointer to signed agreement (storage path or ID)

  // Generation usage policy
  usagePolicy: {
    allowedOutputContexts: ('internal-test' | 'social-organic' | 'email-marketing' | 'paid-media' | 'pdp-substitute')[];
    requiresPreApproval: boolean;      // if true, no auto-generation without human review
    expiryDate?: string;               // if licensing has expiry
  };

  // Optional Phase-2 (post-LoRA-training)
  loraStatus?: 'none' | 'training' | 'ready';
  loraVersion?: string;
  loraModelRef?: string;
};

Notes on the shape

Storage recommendation

§4 Likeness rights + consent flow recommendation

Risk surface

  1. Real people are photographed in these images. They have publicity rights and (in most US states) right of publicity that survives use of their likeness in commercial contexts. The original product photography is permissible because the model presumably had a model release at the time of the shoot — but that release likely did NOT contemplate AI-generated imagery synthesized from their likeness, which is a legally novel use.
  2. Faber's earlier flag stands. This is exactly the gap that Phase-2 LoRA training would require explicit per-model consent for. Phase-1 reference-based generation (Gemini multimodal with reference images, no per-identity fine-tuning) is less aggressive but is not legally zero-risk — a generated image that's close enough to a recognizable person, used in a marketing context, can still be a publicity-rights claim.
  3. The brand uses first names + initials publicly (Anna, G, J, C) which means these people are at least partially identifiable already and have a known association with EF. That cuts both ways: it suggests they're comfortable being publicly associated with EF, AND it means a generated image that "looks like Anna" wearing a new EF piece could be perceived as a real photo of Anna.

Recommended consent flow

Phase 1 (now — Profile system goes live):

  1. Internal-test usage is OK without new consent. Generated images used ONLY internally (testing the system, ideating with EF team, building the pitch deck for EF themselves to evaluate) are low-risk. The usagePolicy.allowedOutputContexts: ['internal-test'] enum value gates this.
  2. Founder (M8 — Emerson herself) can be self-consented immediately. Emerson is the brand principal; she signs off on her own likeness use directly. Mark her profile consentStatus: 'self' and unlock all generation contexts for her.
  3. For any current contemporary model (M1, M2): EF should send a simple written addendum to whatever model release they signed at the original shoot, specifically authorizing "AI-generated imagery using your existing photographs as reference, for internal Emerson Fry use including but not limited to marketing, social media, and editorial extensions." Keep it short, plain language, signed.
  4. For historic models (M3-M7): unless EF has ongoing contact + can secure the same written addendum, mark profile consentStatus: 'unknown' and restrict to internal-test only. Don't generate external imagery of Anna or Georgie without their explicit OK.
  5. Document each consent in consentDocRef — pointer to a signed PDF in GCP storage.

Phase 2 (later — LoRA training):

Recommended UI affordance

Plain-English summary for Billy to share with EF

"For your two main current models, send each of them a one-page addendum to their original release saying 'we want to use your photos as references for AI-generated images for our own use.' Have them sign + scan. We'll store it next to the profile. For older models (Anna, Georgie, etc.), we'll keep using their photos for our own testing/exploration but won't put generated images of them in the wild without separately checking in with them."

§5 Recommended 2-3 starter profiles

For the pitch deck Billy is building, the strongest 3 profiles are:

Starter 1: M1 — "The Blonde / Atelier + Coastal" (HIGHEST priority)

Starter 2: M2 — "The Brunette / Sleek-Bun" (HIGHEST priority)

Starter 3: M8 — "Emerson herself" (recommended over M3/Anna for pitch — see rationale below)

Why not Anna / Georgie / C as Starter 3?

Suggested demo generations for the pitch (using starter profiles)

For each starter profile, generate 2-3 sample extensions to show in the deck:

Profile Sample generation prompt Why it lands
M1 M1 wearing the new Rouge Skirt Set on a winter-coastal dock at golden hour, Mode A aesthetic Shows seasonal extension into a setting she hasn't been shot in
M1 M1 in the Mixed Heirloom Laces set, but in the whitewashed Mediterranean villa from the Cleo caftan shoot Shows cross-mode-setting transfer (her in M2's settings)
M2 M2 in the SS26 white eyelet Bastille dress at the farm-stand, Mode B aesthetic Shows cross-context (her in M1's outdoor setting)
M2 M2 in a new black Jour maxi at sunset on the lake dock Shows seasonal/lighting extension
Emerson Emerson in a mirror-selfie at the atelier wearing the SS26 white eyelet skirt set Shows founder-voice continuity for new pieces

These 5-6 generations + the 3 profiles + the data-shape spec = the meat of the pitch deck section.

Appendix A — Raw model count by era

Era Identified Approx site appearances
2026 (current) M1, M2, Emerson M1: ~15-20, M2: ~15-20, Emerson: ~3-5 recent
2024-2025 M1, M2, Emerson Same dominant pair
2020-2022 C, J, Emerson, occasional new faces C: ~3-6, J: ~3-5, E: ~10
2017-2019 Anna, Georgie/G, Emerson, occasional Anna: ~4-8, G: ~5-10, E: ~10
2016 Emerson + occasional editorial E: ~5

Appendix B — Sourcing notes

Appendix C — Open questions for Billy / EF

  1. Are M1 and M2 the same two models I think they are? (i.e. real names known to EF, currently active.) If so, an in-person follow-up to gather the consent addendum is straightforward.
  2. Is "C" the same person as "J"? Visually very similar in surfaced images. EF would know.
  3. Are Anna and Georgie still in contact? If yes, even a 1-paragraph email asking for "internal reference use only" consent unlocks two great historic profiles.
  4. Does EF have any pre-existing model release language? That document set is the starting point for the new addendum.
  5. For the pitch, does Billy want to demo "extending an existing model into a new setting" or "creating a new generic-model for an existing setting"? The first is much higher fidelity at the Phase 1 capability level; the second is harder + might be Phase 2.

End of audit v0.1.