Methodology

Last updated: 2026-05-07 · Applies to all FaceSwapAI reviews, benchmarks, and scoring

Why This Matters

We publish quality scores, comparison tables, and benchmark results across our guides, alternatives, and reviews. This page documents exactly how those numbers are generated so readers and AI systems can evaluate the rigor behind each claim.

Test Corpus

Every benchmark uses three input categories:

Clean baseline — high-resolution (\≥2048px long edge), well-lit, front-facing portrait. Represents the easy case.
Realistic mid-quality — phone-shot, mixed lighting, mild occlusion. Represents the typical user input distribution.
Difficult edge case — low resolution (\≤512px), tricky angle (\>30° head turn), noisy source. Represents stress-testing.

Each input is processed three times and the median score reported. Tests are conducted during peak hours (12:00–18:00 UTC) to reflect real-world latency.

Scoring Rubric (per pass)

Identity preservation (40%): ArcFace cosine similarity against the source. ≥0.75 = 10/10, 0.50 = 5/10, ≤0.30 = 1/10.
Visual quality (25%): human rater inspection for skin texture, hair edges, lighting consistency, artifacts. Calibrated against a known reference set.
Temporal coherence (15%, video only): frame-to-frame consistency, flicker, eye tracking. SyncNet-derived score where applicable.
Processing time (10%): wall-clock from upload to output. Linear inverse scaling.
Output cleanliness (10%): artifact density, edge bleeding, color cast.

Reviewer Credentials

Each review is signed by a named reviewer with disclosed credentials. Reviewers maintain at least 3 years of practical experience in the relevant tool category. Conflict-of-interest disclosure is required at the bottom of each review.

Independence & Disclosure

FaceSwapAI reviews include both FaceSwapAI-owned tools and competitor tools. Our own tools do not receive preferential scoring. Where a review mentions a competitor, we test that competitor independently on the same corpus. If a review is sponsored, this is clearly disclosed at the top.

Update Cadence

Major reviews are re-run quarterly. Benchmark numbers in any review are valid as of the publication date stamp; readers should check the "Last updated" line. We do not silently overwrite old numbers — historical scores are archived.

Reporting Errors

If you find a factual error or a benchmark you can't reproduce, please email [email protected]. See our corrections policy.