We can build the tool. What it should be depends on what we sell.

We saw all four toolsets on this trip. Nothing there is beyond us technically. The difference is the spec, and the spec follows from direction.

The question is not whether to build our own tool. It is what the tool is being built for. Until the purpose is defined there is no spec.

1. 1Q1 · Quality No secret sauce; the difference is the skills The process is the same as ours. Everyone reaches the same models and the shell is not hard to build. The ceiling on quality is the judgement written into the skills. (pages 4, 6)
2. 2Q1 · Quality Quality is set by the script, not the picture All four put their heaviest effort into the script stage, which is exactly why that stage should not leave the building. (page 14)
3. 3Q2 · Volume Their volume comes from headcount, not from the tool Everyone’s heaviest strategy is the same: hand production capacity out to more people. iQIYI to outside creators, Wawayu to one-person studios, Jiuzhou to newcomers. (page 8)
4. 4Q2 · Volume Cost attribution has to come before volume More than eight in ten are discarded, and the unit cost of those discards decides the margin. Every generation must land on a named person. (pages 9, 10)
5. 5Q3 · Overseas The opening is the marginal cost of extra languages Huace’s localisation centre covers over 50 languages with capacity above 1,000 hours a year; Jiuzhou’s translation hub handles over a thousand titles a month. The back end is cheap, so extra language versions pay. (pages 13, 15)
6. 6Across all three Use someone else’s tool and the bargaining power sits with them And the spec was always derived from strategy. Until we settle what we intend to sell, the tool has no spec. (pages 5, 7)

So four things have to be settled first

1. 1Polished, or fast short drama Huace does premium, Jiuzhou does volume, and both make money. The difference is how they monetise. Settle monetisation and the quality route follows.
2. 2Push for volume, or make something that represents us
3. 3A content factory for internal products, or the leading short-video player in the region
4. 4Invest harder, or hold cost down

The six conclusions map onto the three questions we went in with; the four items are what we need decided. The next page answers each question in full, and the pages after it show how we got there.

The three questions we went in with

We had already built the full pipeline within a month and confirmed it produces video to a standard process; Champion has verified it runs. These were the three open questions we took with us.

All three answers point the same way: quality rides on the skills, volume rides on headcount, and overseas rides on the marginal cost of extra languages. None of the three is solved by buying a tool.

Four routes, four sets of numbers

One industry, four routes. A different business sits behind each number, so each asks a different thing of its tool.

“Last year with live action we did about 200 a month. As of this year we can put up 200 in a week. That is entirely down to the technology, combined with AI.” Jiuzhou Culture · translated

No secret sauce, just practice

A novel or premise comes in; script breakdown, shot breakdown, generation, editing, dubbing, publishing. The order, the stages and the places you get stuck are the same as ours today. They hold no method we do not know. The difference is that they have run these stages many times and fixed the way they do them.

The pages that follow set out where the real difference lies.

So what holds us up will not be the technology. It is that the purpose is still undefined.

Tool specs are derived from strategy

All four tools look different because all four businesses are different. The spec is not set by the technology. It is set by the business.

So our spec has to be derived from what we intend to sell. Until that is settled, there is no spec to derive.

The core is the skills, not the tool

Everyone can reach the same models, and the shell around them is not hard to build. What these four actually cultivate, and never release, is the skills inside the tool.

The senior staff who first wrote Jiuzhou’s judgement into the tool have left. The tool stayed and still runs with different people operating it. The judgement settled into the skills, not into the people.

“The one thing nobody shares is how the skills inside the tool are written. That is the only value these tools hold.” Huace Media · translated

So the question is not whether to build our own tool. It is who writes the skills, and who they are written for.

Four risks to weigh before using a third-party tool

None of the four is unacceptable, but we should know what we are accepting, and which part must not go out with it.

All four are handing production capacity outward

Everyone’s heaviest strategy right now is the same one: let more people produce. They differ only in who they hand it to.

What they are scaling is not output. It is the number of people who can produce. If that is not what we intend to scale, our spec looks completely different.

Credits are how they hold generation cost down

Each team gets a pool of credits, and credits are what generation costs. Spend and revenue feed straight into the bonus.

“Inside our teams the credit allocation, the monitoring, looking at output per head, running the team — it all helps, and it helps a lot on cost.” iQIYI · translated

This spec is independent of what we sell, so doing it early is never wasted.

Fewer than two in ten get traction

“In this industry, out of a hundred titles tested maybe twenty are stories that break out. Not every story we ourselves rate will break out. Twenty per cent — for our house that counts as very high in this industry.” Jiuzhou Culture · translated

Traditional television production tolerates far less failure than this. Taking this route requires agreement with finance first.

If we are going premium, this spec inverts. Settle the direction first, then we know which one we need.

The creator supply is built on policy subsidy

Both Wawayu and Huace run their own training and industrial parks. The policy detail below is public information gathered after the meetings, not something said on site.

Setting up parks and training institutions collects the subsidy, trains the creators and secures the content. All three are one design.

A tool only covers that part. Creator supply has to be solved separately.

The method transfers, the headcount does not

Any plan premised on acquiring thousands of creators is not viable in Taiwan.

So the spec is about output per person, not total headcount.

Before deciding whether to build, look at what the tool amplifies

None of the four earns its revenue from the tool itself. The tool amplifies an existing asset. So we first have to be clear on what we would be amplifying before build-or-buy means anything.

Without an existing asset to amplify, the return on introducing a tool is limited.

So the answer is not in the tool. It is in what we intend to amplify. Until that is settled the tool has no spec.

Keep the front end, hand out the back end

There are plenty of tools that generate video and plenty that check its quality. Early on, using third-party tools for speed is fine. What should not leave is the script and the creative work in front of it.

This is the most concrete split we have seen. Whether we take it still comes back to the four items on the last page.

The only platform in the room told us to run a platform

bilibili is the one company of the five that does not produce its own content. Over dinner we asked which model AI short drama should take overseas.

“For a company like TVBS I think running a platform is the better play. If it is only content production, you can make money but not that much money, so it does not mean a great deal.” bilibili · translated

What we can deliver inside a month

None of this depends on the direction being settled, so it can start now. Only by producing a batch will we know where the workflow bottlenecks and whether quality holds.

Once that batch exists, the four items on page 1 have something concrete to be argued over.