# One presenter. Two believable scenes.

Two UGC-style stills with a continuity sheet, not invented customer testimony.

## What you need

A fictional adult character reference or a consenting adult's approved reference; a product photo you may use. This exercise creates still images. Video and voice require separate tools and permission.

## The steps

1. Write a continuity sheet: facial cues, hair, clothing, jewellery and product grip. Separate identity from background. Use the existing fictional course presenter as a reference if you need a practice input.

2. Attach presenter and product separately. Run scene A, then scene B with the same original references. Limit the exercise to one attempt per scene before reviewing.

3. Compare face shape, hairline, garment seams, hands and product size side by side. Log drift instead of claiming consistency because the same prompt was used.

4. Label shared results synthetic. Use neutral demonstration copy; never put a fabricated purchase experience, health result or endorsement into the presenter's mouth.

## Prompt

```text
Create a vertical UGC-style still, not a video.
Reference A defines the adult presenter. Reference B defines the product.
Preserve: [continuity sheet].
Scene: [A: bright kitchen counter / B: sheltered café table; choose one].
Action: holding the product beside the body, fingers visible, label facing the camera without bending it.
Expression: relaxed closed-mouth smile. Camera: casual phone-level framing, natural skin texture and soft available light.
Do not invent logos, product results, testimonial text or additional people. Keep enough empty space for a synthetic-presenter disclosure added in an editor.
```

## Review

- [ ] Both stills match the original identity and product references.
- [ ] Hands and grip are physically plausible at full size.
- [ ] Synthetic status is disclosed; no customer experience is implied.

## If it fails

Simplify the pose before adding more descriptors. A product resting on the counter is a useful fallback when a hand-held version distorts it.

## Recall later

Does keeping the same name in a prompt prove you kept the same face?

No. Compare visible identity cues against the original reference on every output.

Tool: https://chatgpt.com/
Written by The Leverage Club. Reference edition 2026-09-08.

## Evidence and limits

Original first-look exercise, not a club benchmark. Earlier course images are practice inputs, not Images 2.5 outputs. Check current access and cost.

Launch: https://openai.com/index/introducing-chatgpt-images-2-5/
API guide: https://developers.openai.com/api/docs/guides/image-generation

## Your run log

Date:
App and exact model label:
Input filenames and permission:
Requested change:
Accepted or rejected, and why:
Elapsed time:
Displayed cost (or unknown):
Next action: