What “good enough to take forward” actually means

The decision people get wrong

At the end of exploration you have sixteen frames and have to choose. The instinct is to pick the one with the fewest visible faults. That is the wrong criterion, and it costs whole afternoons.

The right question is: which faults can repair fix, and which can it not? Choose on the second category and ignore the first entirely.

Cannot be fixed by repair

These decide the choice, because fixing them means generating again:

• Composition. Where things sit in the frame, the balance, how the eye moves. Inpainting works region by region; it cannot restructure a picture.
• Light direction and quality. You can grade colour and lift shadows. You cannot move a light source. A flatly lit frame stays flat.
• Pose and gesture. The whole attitude of a body. Repairing a badly posed figure means replacing it, which is a new generation.
• Perspective and camera position. Baked into every part of the frame at once.
• Whether the mood is right. Sounds soft, but it is the most expensive to be wrong about, because it is why the client will reject it.

Can be fixed, so ignore them now

• Hands. Always fixable. Mask and regenerate, or composite. Never reject a good composition for hands.
• Faces at small size. Fixable with a face-targeted repair pass.
• Small objects, jewellery, buttons, cutlery. Inpainting.
• Text. Never generated; always set afterwards.
• Background clutter. Removal or regeneration of that region.
• Colour grade, contrast, warmth. Editing, at the very end.
• Resolution and sharpness. Upscaling.
• Minor anatomical oddities. Repair.

The checklist

Run it on your two or three marked frames at 100%:

1. Does the composition work at thumbnail size? (if no, discard)
2. Is the light directional and consistent: do all shadows agree? (if no, discard)
3. Is the pose right, without repair? (if no, discard)
4. Is the crop-safe core intact for every required ratio? (if no, can outpainting add room? if not, discard)
5. Are there structural impossibilities: a limb emerging from the wrong place, a chair with three legs, a wall that does not meet the floor? (if yes, discard)
6. Everything else: list it as repair work and estimate the minutes.

If a frame passes one to five, it is a keeper regardless of how rough it looks. If it fails any of them, discard it without negotiation, no matter how good the rest is.

Counting the repair cost before you commit

Item six is not decoration. Write the list out and put a number against each:

seed 418
  hands, both, holding chillies        12 min
  face at 3/4, slight distortion        5 min
  bangle merges into wrist              4 min
  background doorway has odd geometry   8 min
  grade + grain                         6 min
                                       35 min

Thirty-five minutes is acceptable. If the list comes to two hours, go back to the contact sheet: another sixteen frames costs one minute and may hand you a thirty-minute version of the same picture. People almost never do this, and it is the single largest avoidable time loss in the whole process.

The honest tension

This checklist is conservative, and conservative choices produce competent, slightly safe work. A frame that fails item five with a strange impossible geometry is sometimes the most interesting image on the sheet, and the rule says discard it.

So hold the rule for commissioned work with a deadline, where predictability is what you are being paid for, and break it deliberately for your own work, where the strange frame is the point. What you should not do is break it by accident at 4pm on delivery day, discover at 6pm that the composition was never going to hold, and have no fallback. Knowing which mode you are in is the difference between a considered risk and an avoidable crisis.

Always keep a second frame

Take two forward, not one. The second is insurance and it costs almost nothing at this stage.

The reason is that some frames do not survive repair. You mask the hands, regenerate, and the new hands are fine but the wrist now reads oddly. You fix the wrist and the sleeve changes. Forty minutes later the image is worse than when you started and you have lost the version you liked, because you were working in place.

Two protections. Save the base render untouched before any repair, always, so you can restart. And keep a second candidate that you have not touched, so that if the first one unravels you have somewhere to go that is not the contact sheet.

Saying no to a frame the client likes

It happens: they pick the one that fails item two. The useful move is not to argue about taste but to name the consequence: “that one is lit flat, so it will look soft next to the others in the set, and I can’t fix that after the fact.” Then offer the one you would choose alongside it, rendered properly.

Presented with both at full quality, people usually change their minds, because the thing that was invisible on a contact sheet is obvious at full size. If they do not, you have said it once, in writing, and you build the flat one.