record-golden-test

Golden replay visual regression records a deterministic play session to capture input events and timestamped keyframe screenshots. This creates a JSON replay file with PNG references that serves as a baseline to validate UI and gameplay visuals after code changes.

You capture keyframe screenshots by starting a recording session at a target resolution like 1920x1080 with a specified keyframe interval, such as 5 seconds. The recording automatically saves periodic timestamped screenshots alongside the input event data into a JSON replay file.

Yes, you can automate UI visual regression checks in a Unity CI pipeline. By injecting recorded input events from a saved JSON replay, the pipeline recaptures screenshots at keyframe timestamps and uses semantic AI vision comparisons to generate automated pass or fail reports for build validation.

Unity visual regression testing works across different platforms and commits by using recorded input injection to reproduce exact gameplay sessions. This deterministic replay mechanism ensures that the same sequence of events is executed to capture comparable keyframe screenshots regardless of environment changes.

AI vision comparison detects gameplay visual regressions by semantically analyzing newly captured keyframe screenshots against saved golden reference PNGs. Instead of strict pixel matching, it evaluates visual differences to identify broken UI, missing entities, or layout shifts for pass or fail reporting.

The best way to catch broken UI or missing entities after Unity code changes is to record a 30 to 60 second reference playthrough, annotate semantic keyframes, and validate subsequent builds by replaying the inputs. AI vision comparisons then automatically detect layout regressions and missing visual elements.