# Family photo review: one person at a time

QuestStudio, October 23, 2026. The example is fictional and was created with the built-in Codex image tool. Its model identifier was not exposed. One source generation and two independent edits; no biometric score or real-family approval.

## Completed example

Job: remove only an orange watering can, leaving its stool and all four people unchanged.
Source: family-study-source.webp, 1672 x 940.
Broad output: family-study-broad.webp, 1672 x 941.
Narrow restart: family-study-narrow.webp, 1672 x 940.
Both requests used the original source, not the previous result.

| Check | Broad request | Narrow restart | Decision |
|---|---|---|---|
| Watering can and stool | Both removed | Can removed, stool retained | Narrow version meets the main background goal |
| Person 1: silver curls, glasses, blue cardigan | Facial and fabric detail rendered with stronger contrast | Broad features and clothing remain similar; fine detail differs | Review at full size; no unchanged-pixels claim |
| Person 2: beard, navy sweater | New white neckline appears | No new white neckline | Reject broad version for this cleanup job |
| Person 3: dark bob, mustard sweater | Brighter skin and clothing treatment | More restrained presentation; fine texture still differs | A familiar person must approve a real-subject edit |
| Person 4: dark waves, green cardigan | Hair and clothing presentation changes | Closer to source presentation, not exact pixel preservation | Inspect hair, buttons, expression and hands |
| Overall | Prompt allowed multiple unnecessary changes | Better match to the intended background-only request | Keep narrow output as a review candidate only |

The broad instruction deliberately asked for extra changes. This demonstrates scope control, not a fair comparison of model accuracy. Neither output is an approved edit of a real family.

## Blank review

- Original source and permission to use it:
- Intended use / audience:
- Exact number of people:
- Left-to-right map (add front/back row where needed):
- One allowed change:
- Model/mode and date, if known:
- Exact prompt and selected format:
- Candidate filename / actual dimensions:

| Person / position | Source anchors | Face and expression | Hair / glasses | Clothes / hands | Decision and reason |
|---|---|---|---|---|---|
| 1 | | | | | |
| 2 | | | | | |
| 3 | | | | | |
| 4 | | | | | |

Use Pass, Revise or Not verifiable. A hidden detail is not a pass. Add rows for everyone, including small faces near the frame edge.

- People counted once each, in the expected positions:
- Unrequested changes to other objects or event details:
- Person familiar with each subject approved the edit:
- Keep / revise from original / stop generation:
- Final filename and sharing disclosure, if a composite:

Keep the original. For a small background cleanup, a conventional edit that leaves the people untouched may serve the task better than regenerating the group.