Tutorials
Claude Opus 5.5: diagnose frame-rate changes that alter ad timing
Give Claude Opus 5.5 the source timestamps, sequence settings, export settings, and measured sync points before asking it to fix video timing.
Give Claude Opus 5.5 the source timestamps, sequence settings, export settings, and measured sync points before asking it to fix video timing. “Make everything 30 fps” is incomplete: interpreting existing frames at a different speed and resampling frames onto a new timeline can produce different durations.
An Adobe community FAQ on variable frame rate documents editing concerns around VFR sources. It is historical guidance, not evidence that a particular current clip is broken. This article uses a calculated example and a proposed Opus review, not a model or export benchmark.
Classify the timing symptom
| Observation | Compare | Avoid assuming |
|---|---|---|
| Same offset at start and end | Stream start times and edit offset | A frame-rate mismatch caused it |
| Offset grows through the clip | Duration, timestamps, speed interpretation | Moving the audio once will fix it |
| Sync stays correct but motion stutters | Source cadence and output frame decisions | The audio needs stretching |
| Only a player disagrees | Another player and editor timeline | The media itself is conclusively wrong |
Keep the original file and record one visible-and-audible event near the start, middle, and end. A clap is useful only when its visible impact and sound really correspond. A music beat over unrelated product motion is not a reliable sync reference.
Calculate what changing interpretation does
Assume a fictional constant-rate clip contains 900 frames at exactly `30000/1001` fps. Its duration is `900 × 1001 / 30000 = 30.03` seconds. If the same 900 frames are instead interpreted at exactly 30 fps, they last 30 seconds: a 0.03-second reduction. For 18,000 frames, the durations are 600.6 and 600 seconds, a 0.6-second difference. These numbers are calculated, not observed in an exported ad.
That calculation applies to reinterpretation of the same frames. It does not mean every conversion from approximately 29.97 to 30 fps changes duration. A timestamp-preserving conversion can duplicate or drop frames to keep the timeline length, with endpoint rounding to inspect. FFmpeg's fps filter describes constant-rate output through frame duplication or dropping. A VFR input needs its timestamps examined; a displayed average alone cannot reconstruct its cadence.
Copyable Opus diagnosis prompt
Use the official Opus guide with the following input contract:
Review the supplied media report, timestamp sample, sequence/export settings, and three measured audio-video sync points. Preserve the original files.
Separate fixed offset, progressive drift, cadence judder, and player-specific behavior. State whether evidence supports each diagnosis. Distinguish frame-rate reinterpretation from timestamp-preserving resampling. Use exact rational rates when available, not rounded labels alone.
Calculate expected duration from frame count only for a confirmed constant-rate stream. For VFR, identify missing timestamp evidence. Do not infer a defect solely because average and nominal rates differ.
Propose the smallest reversible test export to a new filename. Specify what should remain unchanged: intended duration, audio speed, resolution, and product motion. Explain any frame duplication or dropping. Return settings to inspect, predicted consequences, and a start/middle/end acceptance table. Mark export and listening checks pending until actual outputs are supplied.Verify the exported file
Check the new file's duration and stream start times, then compare the three sync events. Inspect a fast-moving product edge or camera pan for repeated-frame judder. Watch the final second: a correct middle does not prove the last caption and CTA survived trimming.
If progressive drift remains, return to the timestamp and speed evidence before stretching audio. If the problem is a constant offset, a targeted alignment adjustment may be more appropriate. Save the accepted settings and output filename with the measured offsets. Do not overwrite the source merely because one preview looks correct.
Delivery questions
**Is VFR always unsuitable?** No. Diagnose the actual editing and delivery path.
**Are 29.97 and 30 interchangeable labels?** No. Record the exact rational rate when available.
**Does constant frame rate guarantee smooth motion?** No. Source cadence and conversion choices still matter.
Prepare approved product imagery in Panelly Studio, then validate timing in your actual video editor. For a problem confined to a repeated ending, use the separate loop-seam guide rather than changing every clip's frame rate.
