Editing: the world is built by what you delete

Generation is cheap. Curation is expensive. This is the fundamental that decides whether your work compounds.

By Ivan Flugelman · 10 min read · Updated 2026-08-18

This is the most important skill you will develop, and the only one that gets harder as the models get better.

When output was scarce, the skill was making something good. Now output is unlimited and the skill is deciding what deserves to exist. Everyone can generate. Almost nobody can delete.

This is the Rejection Matrix, and it is the mechanism by which a world becomes a world instead of a folder.

Kill your darlings

The hardest images to cut are not the bad ones. Bad images are easy. The hardest are the technically excellent images that belong to a different world.

Every time you keep one, you teach your audience that the rules are negotiable, and the world loses a little of its gravity. Ten such decisions and there is no world left, only a portfolio.

The question is never “is this good?” The question is “does this belong?” Those are different questions and only the second one builds anything.

Without rejection, you get a feed. With rejection, you get a world.

The curation ratio

A working session, in realistic numbers:

• Generate 100 images.
• Select 20 that are technically strong.
• Keep 5 that belong in the world.
• Feature 1 or 2 that define it.

If your ratio is much softer than this, you are publishing a feed. Five percent is not pessimism, it is the price of coherence, and it is the same ratio any working photographer or editor would recognize.

What to cut

Concretely, these go, regardless of quality:

• Images where the palette drifts. A warm accent in a cold world, a blue tone in an amber world.
• Images where the lighting does not match the established atmosphere. Hard shadows in a soft-light world.
• Images where the emotional register shifts. A joyful image in a melancholy world.
• Images that are impressive alone but do not sit well beside the others. The great image, wrong world problem.
• Images where the composition breaks the archetype. A centered shot in an asymmetric world, a close-up in a monumental-scale world.
• Images where the materials do not match. Glass and chrome in a world of stone and moss.
• Anything that makes you think “this is great, but.” The but is the world telling you it does not belong.

The grid test

Never judge an image on its own. Put twenty of them in one grid at thumbnail size and look at the grid, not the images.

At that size, individual quality becomes invisible and coherence becomes obvious. Palette drift jumps out. Tonal breaks jump out. The one image that belongs to another world is unmistakable, and it is usually the one you were most proud of.

This single habit does more for a body of work than any prompting improvement. It is also how you find out whether you have a visual language at all, because a world with a real system reads as one texture at thumbnail scale.

How editing connects

Editing is the enforcement layer for every other fundamental. Run the image against each:

• Thesis. Does it serve the argument? If not, cut it.
• Internal logic. Does it obey the laws? If not, cut it.
• Visual language. Correct palette, materials, textures? If not, cut it.
• Tone. Does it match the register? If not, cut it.
• Pacing. Does it serve the sequence? If it disrupts the rhythm without purpose, cut it.
• Archetypes. Does it reinforce the vocabulary, or introduce one that does not belong?

What this looks like in practice

A realistic audit of forty images from one session in a quiet, still world:

• 8 with full adherence to visual language, tone and thesis.
• 12 close, with minor inconsistencies. A color slightly off, a composition too dynamic for a still world.
• 15 technically strong that do not feel like this world. They could belong to any quiet world, not specifically this one.
• 5 that clearly do not belong. Wrong palette, wrong energy, wrong scale.

The interesting group is the third one. Fifteen good images that are generically correct is the most common failure in AI work, and it has nothing to do with skill. It is what happens when a world has a look but not laws.

Common mistakes

1. Judging images one at a time at full size. Coherence is only visible in a grid.
2. Keeping an image because it took effort. Sunk cost is not a world rule.
3. Editing images instead of cutting them. Fixing an off-world image usually produces a compromised image that still does not belong.
4. Publishing everything technically strong. That is the definition of a feed.

Practice

1. Take your last forty generations, grid them at thumbnail size, and sort them into the four groups above. Count honestly.
2. Cut to five. Publish two.
3. Then write your own rejection list, in your own words, for the specific ways your world tends to drift. Yours will not be the same as mine.

The method is free. The practice is the Academy.

Everything above is the whole method. The World Building Academy is the part a page cannot carry: the fundamentals on video, a full world built end to end, the Render Stack, motion, editing, and the case studies. Or start with the free Codex.

Read next