Foundation
Connections clues, built for sharper AI images.
Sort 16 inspection terms into four groups of four. Test a theory, get a near-match nudge, or reveal help one layer at a time without dumping every answer at once.
Practice board / daily edition
Find four related inspection terms
Reveal only what you need.
Start with a broad hint. If that is not enough, expose one term at a time or reveal the complete group. Using assistance is recorded only in your local result; no account or public leaderboard is involved.
Detail
Connection not revealed
Technical
Connection not revealed
Context
Connection not revealed
Strategy guide
How to solve the board without guessing blindly
The puzzle rewards the same habits used in image review: classify evidence, compare alternatives, and protect limited attempts until a theory explains the whole board.
Start with concrete anatomy
Visible face parts such as jaw, ear, lips, or eyelid are easier anchors than abstract terms. Build around the least ambiguous tile first.
Separate cause from symptom
Noise may come from compression, while smoothing may come from generation. Ask whether each term names a defect, a location, an input condition, or a review action.
Test the leftovers
Before submitting four plausible terms, remove them mentally. If the remaining twelve become harder to classify, your first group is probably too broad.
Use near-match feedback carefully
One away means three selected terms belong together, not that the fourth has an obvious replacement. Rebuild the set instead of cycling random tiles.
Protect context terms
Consent, source, metadata, and caption can feel less visual, but they form essential publishing checks. Do not force every tile into a pixel-level category.
Switch to Plan mode
When two groups overlap in your head, place tentative sets in the scratchpad. Seeing four complete alternatives often exposes the single misplaced term.
Common traps
What usually wastes a mistake
Grouping everything visible
Four terms can all appear on a face without sharing the intended inspection purpose. Look for a precise relationship, not simple proximity.
Ignoring workflow actions
Crop, rotate, disclosure, and reverse search describe actions rather than artifacts. Keeping action words together can unlock a difficult board.
Treating AI as the only cause
Compression, camera blur, and lighting can damage an image before generation. Good review distinguishes model failure from source-file failure.
Submitting the same theory repeatedly
After a near match, step back and inspect the other groups. A solved category often removes the decoy that made your first theory look convincing.
Five-pass method
Turn the puzzle into an image-review routine
- 1
Check source quality first: focus, angle, resolution, lighting, and obstruction determine how much reliable information the model receives.
- 2
Inspect boundaries next: trace hairline, ears, jaw, and neck at full size for halos, doubled edges, and abrupt texture changes.
- 3
Compare paired details: eyes, cheeks, reflections, and shadows should follow the same focus and light direction.
- 4
Separate generation defects from compression or export damage so you fix the right stage of the workflow.
- 5
Finish with context: confirm permission, disclose the edit, preserve the source, and avoid presenting synthetic content as documentary evidence.
Questions and answers
Face Swap Connections FAQ
How does the daily puzzle work?+
Each UTC day selects one group from four difficulty pools. The resulting 16 terms are shuffled into a new board. Select four terms and submit; four mistakes end the solve, while Plan mode has no penalty.
Does the puzzle use today's New York Times answers?+
No. Every group and explanation is original and focused on face swap quality, AI image inspection, and responsible sharing. It does not reproduce or reveal any third-party daily puzzle.
What does 'one away' mean?+
Three of the four selected terms belong to one unsolved group. It does not reveal which three. Review the other possible categories before replacing tiles at random.
Are my results uploaded or public?+
No. Streak and completion data are stored locally in your browser. The game does not require an account and does not publish a leaderboard.
Can this game prove that an image is AI-generated?+
No. The terms teach useful review patterns, but a visual clue is not conclusive proof. When authenticity matters, combine source verification, provenance, metadata, and contextual reporting.
Why include consent and disclosure in an image puzzle?+
A technically clean face swap can still be misleading or harmful. Permission, context, and clear disclosure are part of a complete quality workflow, not optional extras after export.
Face Swap Connections is an independent educational puzzle. It is not affiliated with, endorsed by, or a reproduction of The New York Times or its games.
Keep training your eye
Move from words to a real visual inspection.
The Artifact Spotter gives you a fictional portrait with one deliberate face swap defect. Find it in three attempts and learn why that region fails.
