A short shot is not automatically a mistake
video-cut reviews the actual rendered file because even a long planned source interval can contain a few frames of an internal reverse shot. Short-shot and dense-cut reports provide candidates to inspect, not a deletion list.
| Symptom | Record |
|---|---|
| An unrelated face flashes at a join | Possible cause: A neighboring source shot was included First check: Complete shot boundaries in the source |
| A hand action appears only halfway | Possible cause: A reaction or action was truncated First check: Its real beginning and completion |
| The image briefly brightens | Possible cause: Exposure, flash or effect First check: Whether there was an actual cut |
| A person abruptly enlarges or shifts | Possible cause: The crop changes near the join First check: Source shot versus output framing |
| Rapid shots advance an action | Possible cause: Deliberate fast cutting First check: Readability at normal speed |
Repair unintended leftovers or missing information. Keep useful fast cuts rather than slowing everything to a uniform duration.
Original join example: removing a stray shot versus restoring a reaction
This fictional edit plan illustrates the decision; it is not a claim that footage was analyzed.
Source order: customer points at invoice → mechanic explains the fee → two-frame passerby insert → customer takes back the invoice and nods. The intended edit keeps the explanation and acceptance.
| Version | Record |
|---|---|
| The passerby was accidentally included | Repair: Find the complete insert boundary and remove it in the edit plan Verify: Keep the final words and the customer’s real next state |
| The customer’s response was cut to two frames | Repair: Restore the complete take-and-nod action from source Verify: Confirm it belongs after this line, not a different exchange |
| The bright frames were a window flash | Repair: Preserve the action and confirm the cause Verify: Do not remove exposure change as if it were a stray shot |
Remove an unrelated insert; do not freeze it to make it longer. Restore a useful response from real source footage rather than stretching two frames into a pause that never occurred.
Locate, compare source, revise the plan and play again
- Inspect this run’s actual export and record the candidate with surrounding pictures. Planned interval length is not enough.
- Play the neighborhood at normal speed to assess action, dialogue and response. Frame stepping locates boundaries but cannot establish viewing comfort.
- Compare the source: was the short shot already there, or did the edit or crop create it? Keep uncertain origin unconfirmed.
- Repair the complete boundary in the edit plan and export again; do not repair only the rendered file while leaving an obsolete plan.
- Compare before and after using the same picture region and detection parameters, then check words, subtitles, actions and playback speed.
If borders or titles affect detection, inspect the main picture while retaining a full-frame or broader comparison. Shrinking the inspected region to hide a defect is not a repair.
Check sound again after moving visual boundaries. video-cut cleans visual hard-cut edges, then adjusts to reliable sentence/pause windows and checks for cuts inside known speech. Do not remove a final word to lose a stray frame. If complete speech conflicts with the desired picture interval, choose source material again or use a separate picture edit retaining complete audio. Zero short-shot candidates does not establish a finished line.
Fewer detected candidates do not prove a good repair
| Tempting shortcut | Record |
|---|---|
| Flash or dissolve over the join | Why it is unreliable: Hides an unresolved action jump Better repair: Align causal source boundaries |
| Freeze two frames to extend them | Why it is unreliable: Invents a pause rather than restoring performance Better repair: Recover the actual response |
| Extreme zoom at the join | Why it is unreliable: May reduce a detection score while cropping hands or faces Better repair: Keep coherent, readable framing within each source shot |
| Raise the threshold until no report appears | Why it is unreliable: Reduces recall without proving a fix Better repair: Keep comparable settings and playback evidence |
Confirm complete words, no duplicated action, readable expressions and objects, and no subtitle/audio drift after rendering. Detection locates candidates; actual playback determines the result.
FAQ
Does a report with no candidates prove there are no flash frames?
No. Threshold, borders, contrast and overlays affect recall. Check the settings against a known problem passage and still play the rendered video normally.
Should every shot shorter than one second be removed?
No. A short shot may carry a response, detail or fast-cut rhythm. Judge its readable information and relation to the event before keeping, restoring or removing it.
Do it with the skill
Ask video-cut to inspect the actual output for short or dense-cut candidates first. Compare source evidence and recommend keeping, restoring or removing shots without automatically moving cut points. Recheck under the same conditions after repair.
Read the method: Rendered-shot review and repairs · Complete dialogue after visual cleanup