How to review your AI dailies and build a selects sheet in one sitting

You finish a generation session with forty clips in a folder, named by timestamp, and they all look roughly fine. You open the first one, watch it, think it is probably usable, and move on. Two hours later you are in the edit, hunting for that one take where the hand did not melt, and you cannot remember which of three similar clips it was.

The problem is not the footage. It is that you are judging clips while also trying to cut with them, which are two different jobs. Feature productions separate them with dailies: a dedicated pass where nobody is cutting, only rating. You can run the same pass alone, in about forty minutes, and it will save you more time than any prompt tweak.

Do the pass before you open the timeline

Set a rule: no editing software until every clip from the session has a rating. Open the folder in a player that lets you step frame by frame and play at half speed. Watch each clip twice. The first time, watch at normal speed and only note your gut reaction. The second time, watch at half speed and look at the edges of the frame, hands, hairlines, text on signs, and the last second, which is where most generations drift.

Do not delete anything yet. Deleting during review makes you cautious, and cautious reviewers keep mediocre clips.

Use three tiers, not stars

Star ratings invite fence-sitting. Use three tiers only:

- A: cuts in as is. No visible flaw at full screen.
- B: usable with a fix. Trim the head or tail, crop in, cover with a cutaway, or speed-ramp past a flaw.
- C: reference only. Wrong, but it taught you something about the prompt.

If you hesitate between A and B, it is a B. If you hesitate between B and C, play it at full screen on your phone. Flaws that vanish on a laptop often show on a phone held at arm's length.

Build the selects sheet

Open a spreadsheet or a plain text file. One row per clip you rated A or B. Use this structure:

```
Clip file | Scene/shot ID | Tier | Usable in/out (seconds) | Flaw to avoid | Cuts to/from | Notes for regen
scene03_shot02_v4.mp4 | 03-02 | A | 0.0-4.2 | none | 03-03 close-up | keep this lighting wording
scene03_shot02_v6.mp4 | 03-02 | B | 1.0-3.5 | hand warps after 3.6 | cutaway to 03-04 | fix finger count in prompt
```

The in and out points matter most. You are recording the safe window of each clip now, while the flaw is fresh, so the edit never asks you to rediscover it. The notes column is where C clips pay off: write one line on what the prompt did wrong, and carry it into the next generation round.

Check coverage against your shot list

Lay the sheet beside your shot list and mark each planned shot as covered, thin, or missing. Covered means at least one A. Thin means only B clips. Missing means nothing usable. This tells you exactly where to spend your next credits, instead of regenerating whatever annoyed you most.

What you should see

Before: you open the timeline, drag in clips by instinct, and discover flaws mid-cut, then go back to the folder to search. After: you open the timeline with a shortlist, known in and out points, and a clear list of which shots still need generating. The first assembly cut goes faster because every decision about quality is already made, and your regeneration rounds become targeted instead of hopeful.

Try this next

Take your most recent session folder and run the pass today with a timer set for forty minutes. Rate every clip, fill the sheet for the A and B tiers, and mark your coverage gaps. Then generate only what the gaps demand.