Skip to content

AI Erases Tool Skill First. A 40-Year Creative Has Until April to Replace It

A Japanese creator who spent four decades being paid to operate dedicated tools has set a date to stop, and a condition to keep going: turn his advantage into value nobody else can supply. The definition he uses is worth stealing, and so is the material he is giving away instead of keeping.

AI-assisted draft. Reviewed and edited by the Phosphene team before publication.

On this page
A worn wooden tool drawer with a row of old hand instruments laid out in order, one slot left empty, lit by a single window light
Editorial illustration; not a model output.

November 30 is his last working day of the year. December is set aside for preparing 2027. Then comes a three-month window, January through March, and a condition he wrote down in public: if the thing he is trying to build does not take shape in that window, everything he currently does stops in April.

The author publishes under CreativeEdge and does AI drama and AI anime production work in Japan. He is not a newcomer writing a warning post. He spent forty years making things by operating dedicated creative tools, and he has been posting production notes and cost breakdowns from AI video work for a while now. The post announcing the deadline is short, practical, and unusually specific about what he thinks the actual problem is. That specificity is why it's worth reading outside Japan.

What he concluded after two years

On New Year's Day 2024 he declared he was closing his business. Two years later he says refusing to build a business on generative AI itself turned out to be the right call, though he describes it as a grudging decision, and he rates it as a good one.

The second half of the plan did not work. His stated goal after the closure was to use generative AI as hard as possible while creating work that generative AI does not affect. That has not happened yet. So he set an end date, and an emergency project to fall back on.

There is a detail in the framing that most commentary on this subject skips. He is not saying AI took his work. He is saying his current advantage is being gradually erased, and his job is to convert what's left into value that only he can provide before the clock runs out.

The layer that went first

Forty years of operating dedicated tools is a specific kind of career. The software was expensive, badly documented, and took years to learn. Clients were not paying for the software. They were paying because you could drive it, and because the number of people who could drive it was small. Interface mastery was the product.

Generative models removed the interface as a barrier. The interface is now a sentence. Everything that used to sit behind years of tool skill, tracking, keyframing, render settings, export quirks, is either automated or good enough by default. That does not make craft worthless. It moves the price floor, and it does so first on the part of the job that was tool operation.

What's left is upstream of the tool: deciding which job to take, deciding which of thirty outputs is the one, knowing why the client's brief is wrong before they do, and being the person who answers when the result is bad. A model does not decide any of that.

What "make the value personal" actually means

The Japanese term he uses is 属人化, which translates awkwardly. In management writing it usually describes a risk: knowledge and judgment locked inside one person so the organization can't replace them. He flips it. For a solo creator, value that cannot be easily substituted is not a risk, it is the lifeline.

He is careful to rule out the cheap version of this. Personalization here does not mean making your process a black box so nobody can follow it. He defines it as strengthening the value rooted in your own judgment, expression, and experience. Small teams, in his view, need both: a shared way of working, and each person's irreplaceable edge.

That definition is easy to nod at and hard to apply. A usable test: take your last ten deliverables, and for each one ask whether a competent stranger with the same tools and the same brief could produce something equally good in a week. Where the answer is yes, that work is priced by the market, and the market price is falling. Where the answer is no, you have found the part worth protecting. Most people discover the honest answer is "yes" for more of their portfolio than they expected.

The record is the asset

One line in the post is the most practical thing in it. He logs essentially everything in a private production notebook, including the work that failed, because that record becomes data, and the data becomes the engine for fast prototyping.

Read that as a workflow instruction rather than a philosophy. The only dataset nobody else has is your own project history: what you tried, what you rejected and why, which prompt produced which frame, which setting broke a face. A person with four decades of taste and no written record has taste that dies with the project. A person with a mediocre eye and three years of labeled rejections has something a model can be pointed at.

What he is giving away

Rather than sitting on unused work, he plans to release it, probably in December, and says he will announce when the upload is complete.

The main item is VELQION, a five-member AI boyband built as a prototype for an AI idol business. The concept was not adopted, so the material is being prepared for release: every member's reference images, the prompts used, and the music data. His terms are explicit, free for non-commercial use, and he points out one particular use himself. The set works as source material for running five characters at once in Seedance 2.5.

That last detail is the part AI video people should notice. Keeping one character consistent across shots is the standard problem. Keeping five consistent in the same frame, with the same faces from different angles, is where most pipelines fall apart, and it's the kind of thing that takes a week of failed generations to assemble. A labeled pack with references per character is genuinely expensive to build and cheap to hand over. He is also considering releasing an unused character from his anime adaptation who does not appear in the live-action version.

Two motives are visible. The stated one is that unused data has no marginal cost. The other, which he does not claim, is that being the origin of other people's material is a form of positioning that survives a falling price floor. I'd call that inference, not his argument.

The number under all of this

In an earlier production note he put a number on the squeeze, and the number needs its scope attached. The claim is video generation cost only, for a three-minute episode, built by fixing every generation at 8 seconds and assuming a 25 percent success rate, meaning four generations per usable clip. On those assumptions, Seedance 2.5 works out to about 47,500 yen and MiniMax H3 Max Turbo to about 4,300 yen. That is where the "one tenth" framing comes from: a cost estimate for the generation step, comparing two models under a stated assumption, not a measured drop in the total cost of delivering a finished episode.

The direction is still the story. When the generation line of a deliverable drops by an order of magnitude, the price the client pays tends to follow, and the studio that was charging for the old process has to either move up the value chain or shrink. His deadline is a response to that arithmetic, not to a bad month. It is an argument about cost structure, not a measured price collapse, and the difference matters if you are using it to plan your own year.

Where I'd push back

This is one person's plan, not a forecast for the field. His business was AI idol and AI anime services, which is close to the most exposed position you can hold: a product that is visibly model-generated, sold to clients who can now generate a version themselves.

Set against that, the plan he wrote is more disciplined than most. He set a date, defined the test, and named what happens if the test fails. Most creative businesses facing the same squeeze improvise until the money runs out, which makes the decision for them at the worst possible moment.

The audit worth running this week

  1. Label your last ten deliverables by layer: tool operation, pipeline assembly, direction and taste, client relationship and accountability. Count how much of the invoice sits in the first two.
  2. Write down the one thing a client has said they can only get from you. If nobody has said it, that's the finding, not a failure of memory.
  3. Start the log he keeps. Per project: prompt, model, settings, what you rejected. Ten minutes a week, and it compounds into the only dataset that describes your own taste.
  4. Put a date on the decision instead of a mood. A quarter is enough to test whether a positioning idea has buyers.

The value you can defend is narrower than the work you currently do. Finding out how much narrower is cheaper now than in April.

Sources

Share this guide
X LinkedIn