How to decide whether to cut around an AI flaw or regenerate the shot

A three-question routine, plus four edit-side fixes, for deciding when a flawed AI clip is worth saving and when to spend the credits on another take.

By Leeby Shmeeby

You have a take that is nearly right. The framing works, the performance is the best of six, and then at 00:03 a hand grows a sixth finger for eleven frames, or a background figure melts into a doorframe. You do one of two things. You regenerate, spend more credits, and get a take that fixes the hand but loses the performance. Or you keep the take, tell yourself nobody will notice, and then see the flaw every time you watch the cut.

Both habits come from deciding by feel. A flaw feels enormous when you have stared at it forty times, and it feels trivial when you are tired of generating. What you need is a short routine that makes the call the same way every time, so you only regenerate when the edit cannot solve the problem.

How to decide whether to cut around an AI flaw or regenerate the shot

The three questions

Scrub the clip once at full speed, then ask these in order. Stop at the first answer that settles it.

```
1. How long is the flaw on screen? (count frames, not impressions)
2. Is the flaw on the subject or in the background?
3. Does the shot need the flawed moment to tell the story?
```

Under roughly half a second, off the subject, and not story-critical: cut around it. Longer than a second, on a face or hands in focus, or on the beat the scene is about: regenerate. Anything in between is a judgement call, and the fixes below decide it.

Four edit-side fixes to try first

Shorten the clip. If the flaw sits in the first or last second, trim it off. Most AI clips are softest at the head and tail anyway, so you lose very little.

Cover it with a cutaway. Place a reaction shot, an insert of an object, or a wide from your coverage set over the flawed frames, and keep the original audio running underneath. Four or five frames of cutaway is usually enough to hide a glitch.

Speed ramp through it. Select the flawed span, speed it to 300 percent, and ramp back to normal either side. The flaw passes in a couple of frames. This works best on movement shots, such as a walk or a pan, where a ramp looks like a stylistic choice.

Mask and patch. For a static background glitch, duplicate the clip onto the track above, offset it to a clean frame, and draw a soft-edged mask over the problem area. Hold the patch for the length of the flaw. Do not try this on anything that moves across the frame.

Applying it to a real take

Say your best take of a woman opening a door has a warped background figure from 00:02 to 00:03. The flaw is in the background, it lasts about a second, and the story beat is her hand on the handle. Trim the head to start at 00:03, and your flaw is gone with no regeneration. If the flaw were on her hand during the handle grab, question three settles it: regenerate, and keep this take as a reference for the performance you want to match.

Log the decision in your take sheet in one word, either saved or rerolled, with the fix used. After a week you will see which flaws you reliably save and which you never do, and the routine gets faster.

What you should see

Before, your rejection pile holds takes that were fine apart from one second, and your credit spend climbs on shots that were already close. After, you reject only takes with flaws the edit cannot hide, and the usable takes in your timeline carry cutaways and trims you chose on purpose. You should also notice that the performance you keep is better, because you stop trading it away for a clean background.

Try this next

Open your current project, find the three takes you rejected most recently, and run the three questions on each. If any pass, try a trim or a cutaway before you generate anything else.