Lately, as our team has been creating AIGC (AI-generated content), I have often run into a problem that rarely came up before. It is not that we cannot make something. It is that we can make too much of it at once. The same shot can yield more than a dozen versions; the same character can take on dozens of states; the same line can be placed in different settings, lighting, compositions, performances, and rhythms. In the past, the hardest part of making content was often a lack of something. No budget, no actors, no locations, no time. Now, increasingly, the real question is: Which one do we use? Over time, I have come to feel that as production gets cheaper, one thing becomes more valuable. Judgment. In the past, conditions made many choices for us. If the budget was insufficient, we did not shoot it. If we could not build the set, we changed the plan. If the actor was unavailable, we rewrote the script. Production capacity itself formed a wall. Generative AI is taking that wall down. But beyond that wall is not a paradise of creative freedom. Beyond it stands another wall: Do you actually know what you want? That is why I use the phrase “AI makes creation easier” less and less. It has made production easier. But it has not necessarily made creation easier.
When reviewing content within our team, the line I least want to hear now is: “Everything is pretty good.” Because when all ten versions are “pretty good,” it usually means we do not actually have a standard. AIGC is especially good at creating this situation. The images are all beautiful. The characters are all polished. The shots all have a cinematic feel. Taken on its own, each one seems to have no problem. But once it enters the work, the questions appear immediately. Why this shot? Why should the character have this expression at this moment? Why this lighting? Does it serve the emotion, or merely display what the model can do? That is why we have come to repeat one line within the team: Generation only gives us material; creation makes an image serve a specific purpose with precision. “Looks good” is not the final standard. Looking good does not mean being useful. Still less does it mean it should be kept.
This is also where AIGC can become most addictive for creators. The model is always willing to keep offering answers. Not satisfied? Generate again. Still not satisfied? Try another prompt. Switch models. Add a reference image. Make ten more versions. In the past, creators’ greatest pain was having no choices. Going forward, the opposite may be true. Too many choices. Yet our time has not increased. Neither has our attention. A user still has only 24 hours in a day. So the truly scarce capability in the future may no longer be “generating more,” but “deleting more.” In an earlier piece about AIGC-native IP, I wrote: Tools are responsible for producing answers at scale; creators are responsible for continuing to say no. I believe that line more and more. A truly mature creator may not be the person who is best at prompting a model. They may be the person who can face a hundred results and say, very quickly: Not this one. Not this one either. This one is beautiful, but it is not right. This might perform well in the metrics, but it is not what we want. Then keep one. And know why they kept it.
That is why my understanding of “taste” is now somewhat different from before. It certainly includes aesthetic sensibility. Composition, color, rhythm, performance, and sound all matter. But in the age of generation, taste is increasingly a form of comprehensive judgment. What is worth making. What is not worth making. What should appear. What should disappear. When something should be complex. When it should be simple. When the model should keep generating. When it is time to stop. This capability is hard to acquire by memorizing a few prompts. It comes from what you have seen. What you have made. What you have failed at. What you have liked over time. What you have rejected over time. It may even come from the life experience a person has built over the past decade or more. So the stronger AI becomes, the less I believe the overly simple claim that “anyone can instantly become a creator.” Anyone can, of course, generate. But selection still stands between generation and creation.
This becomes even clearer with AIGC-native IP. AI is especially good at creating variation. Today the character wears a suit. Tomorrow, a spacesuit. The day after tomorrow, they become a person from ancient times. In the next post, they appear underwater. Every version can look great. But what IP fears most is change without a reason. Because users need to build an understanding. They need to know: Who this person is. Why they speak this way. What relationship they have with others. How they would choose when they encounter something. If every generation of a character is only chasing “this time, make them cooler,” what often remains in the end is not a character, but a pile of beautiful assets. That is why I increasingly feel: AI is best at creating change; IP most fears change without a reason. At its core, that line is still about taste. Taste is not making everything more beautiful. Taste is knowing when something should not change.
When we used to discuss the industrialization of content, we often emphasized standardization. But since generative AI entered the picture, I think the focus of standardization may change as well. In the past, it was about standardizing production actions. In the future, it may also mean standardizing criteria for judgment. What is ready to publish? What does it mean for a character to drift from who they are? When is an image beautiful but not serving the narrative? When does commercial content harm the IP? When would we rather not publish? If these things exist only in the mind of a particular director, writer-director, or manager, it will be very difficult for a company to truly scale. So one of the most important assets of a mature AIGC content team may not be how many model accounts it has. It may be whether it has a stable system of judgment. That system ultimately determines: Among the ten thousand possibilities AI generates, which ones can enter the work.
Many people ask me what creators should learn most in the age of AI. I am now less inclined to recommend tools first. Because tools change too quickly. A function that is crucial today may become a model’s default capability six months from now. What is truly worth training over the long term is judgment. Watch more work. Make more real projects. Meet users more often. Experience more failures. Know what is merely “technically impressive.” Know what can genuinely make someone stop. Know what you like. More importantly, know why you like it. As a person’s standards become clearer, AI can truly become leverage. Otherwise, the stronger the tools become, the easier it is to be drawn into endless generation.
I even think we may be moving gradually from an era of “production scarcity” toward an era of “selection scarcity.” The old problem was: You could imagine it, but could not make it. More and more, the problem may become: There are too many things we can make, but we do not know what is worth making. These two eras call for entirely different capabilities. The earlier era rewarded execution. The later era may increasingly reward judgment. So AI will not make taste less important. Quite the opposite. When everyone can generate a beautiful image, “beautiful” itself may quickly become less distinctive. When everyone can make a shot that looks cinematic, the cinematic look may become less distinctive too. What remains central is: Why did you choose this?
So if I had to rethink the word “creator” today, I might say: A creator is not the person who can make the most things. A creator is the person who knows what ought to exist. AI will likely become increasingly good at answering “how do we do this?” But there is one question it is difficult for AI to take on for us: What is worth generating? I think that question will become increasingly important. It may even become one of the central creative questions of the generative age. When facing a hundred versions, the valuable capability is not to generate the hundred-and-first. It is to clearly delete ninety-nine. Then keep this one. Because you know why you want it.
Copy this essay for someone thinking about AIGC content, IP, organisation or digital identity.
