Ai Composition Studies #30: The Diplomacy of Revision Costs
The cost of one feedback revision is 50 minutes.
25 minutes in UE5, 25 minutes in After Effects. They can't run simultaneously, so you wait in sequence. One small revision takes an hour. Three rounds of feedback take three hours. Accumulated across all cuts, a day disappears.
This is from the reality of game cutscene lighting. But every production environment has similar numbers.
Revision costs are not a technical problem. They are a diplomatic problem.
Technology makes work faster. Diplomacy eliminates revisions.
The True Nature of Feedback Hell
Feedback hell is not proof of quality pursuit.
It is a symptom of absent judgment.
"Make it a bit softer." "The fog looks inconsistent." "Something's off."
These are not feedback. They are impressions.
When impressions are thrown at workers, they can only guess. They revise based on guessing. They're told it's wrong. They guess again. Wrong again. This back-and-forth creates feedback hell.
Most causes of feedback hell lie on the direction side.
It's not that workers lack skill. Judgment criteria haven't been shared. Without criteria, no matter how many revisions, "wrong" keeps coming. With criteria, it ends in one round.
Two Types of Feedback
There are two types of feedback.
One is feedback from skill deficiency. "Shadow settings are wrong." "Light falloff curve is too steep." "Exposure is 0.5 too high." This is specific and ends in one round. Fix it and it's done.
The other is feedback from absent judgment. "A bit softer." "The atmosphere is different." "Not enough weight." This is vague and goes back and forth repeatedly. Fix and fix, but it never ends.
The difference is clear.
The former points out "what is wrong." The latter is "feeling something is wrong without knowing what is right."
Let me talk about gatekeepers.
Gatekeepers exist to open gates. Pass what should pass, stop what should be stopped. That's the job.
But there are gatekeepers who demonstrate their value by closing gates. Stop without reason. Send back without criteria. No matter how many times you come, they keep saying something is different.
That's not a gatekeeper. It's just an obstacle.
Directors without judgment criteria become obstacles.
The Same Applies to AI Instructions
This isn't just about human-to-human interactions.
Instructions to AI have exactly the same structure.
"Make it better." "Something's off." "Add atmosphere."
With these instructions, AI can only guess. It outputs based on guessing. You say it's wrong. It guesses again. Wrong again. The same feedback hell as human-to-human occurs.
But what if you instruct like this?
"Maintain light direction." "Shift color temperature warmer." "Enrich shadow gradation." "Don't change composition."
Give specific criteria, and AI's output gets close in one round.
Direction to humans and prompts to AI are essentially the same. Make judgment criteria explicit, and revisions decrease. Throw things vaguely, and back-and-forth increases.
The principles of diplomacy don't change whether the counterpart is human or AI.
Crunch = Diplomatic Failure
Crunch is not passion.
It's the result of no one capable of design.
Think in terms of international relations. Resolve through diplomacy, and it costs a conference room. Go to war, and the nation tilts. Costs multiply thousands, tens of thousands of times.
Production is the same.
Share judgment criteria in advance, and it costs one meeting. Proceed without sharing, and feedback hell begins. Revision costs pile up, schedules collapse, teams exhaust.
I've seen crunch projects. All the same pattern.
They didn't design judgment criteria at first. They started running vaguely. "Wrong" was repeated midway. Revisions piled up, and nobody knew the right answer anymore. In the end, it became a stamina contest.
This is war. War brought by diplomatic failure.
Visual Direction Book = Diplomatic Document
There's a way to avoid war.
Create a Visual Direction Book.
This is a diplomatic document. Prior sharing of judgment criteria. What is right, what is wrong. What to aim for, what to avoid. Specific values, reference images, and NG examples.
When I start a lighting job, I create this first.
Standards for light direction
Color temperature range
Contrast upper and lower limits
Shadow hardness standards
NG examples and their reasons
With this, feedback becomes specific. "Shadow is harder than standard." "Color temperature is out of range." Fixed in one round.
Without this, feedback becomes impressions. "Something's off." "A little more." Back and forth repeatedly.
Going to war without creating diplomatic documents is folly.
Reading C from c-
Here I bring in the Ai Composition Studies perspective.
In Ai Composition Studies, c- (c-minus) refers to "connection points that appear on the surface," and C refers to "the compositional core behind them." c- is measurable; C is unmeasurable but exists.
Examples of c-:
Rendering time
Number of feedback rounds
Working hours
Days spent on revisions
These are visible as numbers.
Examples of C:
Absent judgment
Vague criteria
Unclear responsibility
Design failure
By observing c-, C emerges.
When feedback rounds exceed three, it's not a skill problem. It's a sign that judgment criteria aren't shared. When rendering time exceeds half a day, it's not a tool problem. It's a sign that diplomacy has failed.
See the structure behind the numbers. That is the Ai Composition Studies perspective.
Problems don't occur on-site. They occur in structure.
Conclusion
Revision costs are not a technical problem. They are a diplomatic problem.
Feedback hell is not passion for quality. It is a symptom of absent judgment.
Crunch is not the team's lack of effort. It is design failure.
Design judgment criteria. Create diplomatic documents. Avoid war.
And don't forget.
Gatekeepers exist to open gates.
Gatekeepers who demonstrate their value by closing gates are not gatekeepers.
#AiCompositionStudies #RevisionCosts #Diplomacy #FeedbackHell #JudgmentCriteria #VisualDirectionBook #AIPrompts #GameDevelopment #Lighting #EiStudio