# JBNX Studio Model m/004 — Bethany

Fictional adult AI actress for the owner's film work. Built with the AI Model Reference Pack Framework:
https://projects.jbnx.io/framework/ai-model-reference-pack-framework · job file: `reference-pack-job.json` · QA: `qa.json`.

## Source and rules

- **Face baseline:** the owner's own AI-generated image (shared 2026-11-01). Not published; its SHA-256 is in the job file.
- **Age:** adult, early-to-mid 20s. Every image must read as an adult and never younger than the approved face anchor.
- **Build:** a curvy hourglass figure, from a body reference the owner shared. That photo was described in text only and never sent to a provider.
- **Wardrobe:** cream ribbed long-sleeve fitted top, charcoal high-waist leggings, white sneakers. Poses are neutral studio poses.

## How it was built (cheap until the finals)

1. **Pilot, about $0.21:** Topaz 4x on the baseline, then three 1K drafts on Nano Banana 2.1 edit (about $0.044 each): face front, body front and body side. Jev scored them against the baseline. The owner approved the face and body front drafts ("approved", 2026-11-01).
2. **Finals at 4K:** Nano Banana Pro edit used the two approved drafts, each restored 2x, as image 1 (face) and image 2 (body). A three-still gate ran first (face, body, hero anchor), then the other 16 stills: 8 body angles, 5 face turns, 3 expressions, 2 scene frames and the hero anchor, plus a 360 sheet.
3. **Motion:** four Kling 3.0 pro baselines (turn, walk, sit, talk; sound off) with one "bethany" element (face 000, face 045, body 000).
4. **Voice:** ElevenLabs "Lily - Velvety Actress" (premade, in the owner's account), chosen by the agent. Sample line: "Hi, I'm Bethany. This is my reference model for JBNX Design." The voice proof is the talk baseline lip-synced with Sync Lipsync 2 Pro. The owner approved the voice and line on 2026-11-03.

## Likeness

Jev and ArcFace face-ID scored every image against the approved face anchor. The table is in `model.json` under `likeness` and on the page. Full-body shots score lower because the face is small at the scorer's input size.

## Status

The owner approved v0.1.0 and the voice line on 2026-11-03. Spend is in `production-ledger.json`; the cap is $10.