
How FINAL EYE tests AI video
Every test compares AI video models, or versions of the same model, on the same prompt. The formats change from test to test; the rubric and the blind judging stay fixed.
The process
Four steps, every test
Fixed rubric
Prompt adherence /100 plus craft axes, set before any clip is seen.
Blind letters
Clips are judged as A, B, C… The model names are revealed only after scoring.
Frame-level proof
Every defect is timecoded and located in the FINAL EYE comparison app.
Human in the loop
In Test01 P1 the app's automatic scan missed two slight camera pans; the human eye caught them.
Tested via Higgsfield, first generation, no retries.
The rubric
Nine axes
Adherence is scored /100 against the prompt's own words. Execution is the average of axes 3–8. Audio is scored separately.
- 1Prompt & brief adherence/100
- 2Reference fidelity & identity/10
- 3Visual integrity & material credibility/10 · EXEC
- 4Light, colour & art direction/10 · EXEC
- 5Motion, physics & causality/10 · EXEC
- 6Temporal stability & scene continuity/10 · EXEC
- 7Camera & optical execution/10 · EXEC
- 8Performance & expression/10 · EXEC
- 9Audio & audiovisual coherence/10
The tool
The tool
Built in-house for this work; not for sale.
- Synced timeline with up to four playheads on the same frame.
- Wipe, split, overlay and difference views.
- Pixel zoom.
- RGB parade and BT.709 vectorscope with a probe.
- Automatic scan for flicker, frozen frames, drift and legal levels.
- A marker per defect with timecode, severity and axis.
Tools
Tools FINAL EYE uses
Independence
No sponsors on verdicts
- No affiliate links on verdicts or rankings.
- Rubric fixed before each test.
- No vendor sponsors in the same test.
- Any commission is disclosed at the top of the page where it applies.
Free report
The free P1 report
See the method on a real test: the Test01 P1 report, sent by email when you subscribe to the FINAL EYE newsletter.
FINAL EYE