// PERSONA IDENTITY

NAME THE PERSONA.
EMBED THE FACE.

The problem is not “we need another influencer dashboard.” The problem is a named AI persona (content identity, synthetic character, branded face) with no biometric ground truth. HunterRose is an AI persona manager on meltuc.tech: create a persona, upload reference images, approve them, and the app builds a 512-dimensional ArcFace fingerprint (insightface buffalo_l). Canonical embedding is the L2-normalized mean of approved faces. Storage: Supabase bucket hunterrose-references, jpg/jpeg/png/webp, 10 MB. Statuses: Incubating / Warming / Active / Paused / Retired. Archive is soft-delete. Version 0.1.0: CRUD + embeddings live. LoRA train, asset generation, similarity review, GPU spend, job page are schema only until RUNPOD_API_KEY. This landing will not fake a GPU studio. Proof: dashboard counts + persona list + upload + approve/reject.

Premier v0.1.0 ArcFace 512 GPU pending
Open Dashboard Personas How it works
512
ArcFace dims
10 MB
Max reference
5
Lifecycle states
0
Live GPU jobs
// HOW IT WORKS

Five steps that actually exist today

Long-term pipeline is five stages; only 1–2 are implemented. Honesty is the product on this page: do not sell RunPod output you cannot generate.

01
Create the persona
/hunterrose/personas/create: name required, niche, bio, voice, visual style. Slug auto, de-duplicated per workspace.
02
Upload references
jpg/jpeg/png/webp ≤10 MB to hunterrose-references. Gallery on persona detail with per-reference status.
03
Approve to embed
Approve runs buffalo_l, stores face_embedding, bbox, detection_confidence, recomputes canonical mean embedding.
04
Manage lifecycle
Incubating/Warming/Active/Paused/Retired. launched_at on first Active. Archive hides from lists. No hard delete on archive.
05
Stop before the GPU myth
LoRA, generation, similarity gate, spend, jobs: designed, not routed. Missing RUNPOD_API_KEY. Health JSON skips DB.
Open HunterRose
// DASHBOARD

Counts by status — CSS mock

Not Midjourney, not Runway, not a generated-asset gallery. Five recent personas and status tiles.

meltuc.tech/hunterrose/app — CSS mock, not a product screenshot
Incubating—
Warming—
Active—
Paused / Retired—
// REFERENCES

Approve runs ArcFace, reject does not

CSS mock of a reference row. Approving embeds. Rejecting does not invent a face vector.

meltuc.tech/hunterrose/personas/<slug> — CSS mock, not a product screenshot
Fileref.webp ≤10MB
Embedderbuffalo_l 512-d
Canonicalmean of approved, L2
// NOT YET

GPU pipeline is a documented hole

B1-03 through B1-08 pending credentials. Do not paste fake LoRA loss curves.

meltuc.tech/hunterrose/ · blocked — CSS mock, not a product screenshot
LoRA trainschema only
Generatenot implemented
Jobs pagenot implemented
BlockerRUNPOD_API_KEY
// LIMITS

No CSV export. No cron. No GPU.

No CSV/JSON export endpoints, no cron in this blueprint. Bulk-delete personas API max 100. Login wall on app/personas.

// WHAT IT REFUSES

Will not pretend assets exist

Will not generate images without RunPod. Will not train LoRA in Flask. Will not store a fake embedding on reject. Public landing lists no faces.

// WHO THIS IS FOR

Operator managing named identities

meltuc.tech admin building persona ground truth for later generation. If you need an image factory today, this app is not ready.

// FEATURES

Live now vs designed later

If a card says pending, it is pending.

👤
Personas
Name, niche, bio, voice, style, slug, statuses.
🖼
References
4 formats, 10 MB, Supabase storage.
🧬
ArcFace
512-d buffalo_l on approve; canonical mean.
🗂
Dashboard
Status counts + five recent.
🔎
List filters
Status + case-insensitive name; archived hidden.
📦
Archive
soft archived_at, hidden, no hard delete there.
🔐
SWIP
Public landing; work behind login.
❤️
Health
/hunterrose/health JSON, no DB.
⏳
GPU later
LoRA/assets/jobs blocked on RUNPOD_API_KEY.
// GET STARTED

Create a persona. Embed a face.

Login for /hunterrose/app. This page is not a gallery of identities.

Open HunterRose

Requires a MelTuc account. Create one free.

// SAID SLOWLY

What stays true over time

AI persona management — create, manage, and version AI personas with reference images. Health stays cheap. If this page disagrees with the signed-in app, trust the app — the product contract and tests enforce that rule.