Werner Herzog’s recent comments about AI in filmmaking are worth reading for more than the headline value. He was not just swatting at a new tool. He was pointing at a workflow problem that shows up whenever a creative process starts by asking software to imitate someone else.
Herzog’s verdict was blunt: “It’s phenomenally, abysmally stupid.” He also said, “It’s so standardized that AI can step in because there’s not real storytelling.” That is a harsh line, but it maps to a real tension in creative production: the closer a process gets to mechanical imitation, the less it depends on judgment.
For builders, that distinction matters. Tools are useful when they reduce friction around a decision. They are much less useful when they quietly replace the decision itself.
Where imitation breaks the workflow
In filmmaking, the work is not just generating visuals. It is a chain of choices that includes:
- what story is worth telling
- what tone the film should have
- what belongs in the frame
- how pacing and structure should evolve
- how much influence to take from references without copying them
If a team begins by asking AI to produce something that looks like an admired filmmaker’s work, the pipeline can become shallow very quickly. The output may be visually convincing, but the creative reasoning underneath it can disappear.
That is the key thing Herzog seems to be objecting to. He is not only saying the output is bad. He is saying the process is bad because it starts with imitation rather than intention.
For anyone building creative tools, that is a useful failure mode to study. A model can help explore options. It can help with variation. It can even help a team move faster during ideation. But once the goal becomes “make it look like X,” the system is no longer helping with authorship. It is helping with approximation.
Why standardized output feels productive
Herzog’s criticism also explains why AI imitation can feel so tempting in the first place. Standardized output is fast. It reduces ambiguity. It gives people something tangible to react to.
That can be useful in early stages of a project. A rough visual direction can help a team align. A generated mood board can clarify what kind of atmosphere a project is aiming for. But there is a tradeoff: the more specific the imitation, the more the creative team risks confusing resemblance with direction.
This matters because filmmaking is not only about speed. It is about making tradeoffs that preserve authorship.
A workflow that privileges imitation may look efficient, but it can create problems later:
1. It hides weak story choices
If the story is thin, a convincing style can postpone the harder question: is the film actually saying anything?
2. It narrows the range of decisions
The team may stop exploring alternatives once the AI output resembles the reference closely enough.
3. It encourages pattern completion over interpretation
The model fills in familiar shapes, but it does not decide what makes the project distinct.
4. It can make polish look like originality
A standardized result can feel finished even when it lacks a clear point of view.
That is why Herzog’s comment lands on workflow, not just taste. A creative process can become optimized for the wrong thing.
The broader point in Herzog’s own language
This is not the first time Herzog has made a sweeping statement about machines and filmmaking. He has previously said that “a computer will not make a film as good as mine in 4,500 years.” Whether you read that as provocation, confidence, or both, the position is consistent.
He does not seem persuaded that machine-generated imitation can substitute for artistic vision. He also framed the issue as part of a larger internet problem, describing it as “the most common, lowest denominator of what is filling billions and billions of informations on the internet.”
That line is revealing because it connects the complaint about AI to the training data environment that makes imitation possible. A model built from huge amounts of online material can produce something recognizable because it has learned the most repeated patterns. That does not automatically make the result meaningful. It often just makes it familiar.
For creative tools, that is an important design constraint. Familiarity is not the same thing as intent.
Practical lessons for builders
If you are building software for creative work, Herzog’s comments can be translated into a few concrete questions:
Is the tool helping people explore or helping them imitate?
Exploration can be useful. Users can test directions, compare options, and discover possibilities they would not have considered. Imitation, especially direct imitation of a living or historical filmmaker, is a much narrower goal.
Does the tool preserve decision-making?
A good creative tool should support the human judgment that gives a project its identity. If the software makes too many aesthetic decisions on its own, the output may become standardized faster than the user realizes.
Are you optimizing for speed or authorship?
Those are not always the same thing. Speed can help a workflow, but authorship requires space for disagreement, revision, and uncertainty.
Is the output solving the actual problem?
If the real issue is weak storytelling, generating a lookalike version of a famous style is not a fix. It may just make the weakness harder to notice.
These questions matter because many creative products are judged by how quickly they can produce something that looks finished. But a finished-looking result is not necessarily a meaningful result.
A useful way to think about AI in filmmaking
The most productive way to read Herzog’s reaction is not as a blanket rejection of all AI. It is a warning about where AI belongs in the process.
Used carefully, AI can support ideation, organization, and experimentation. Used badly, it can flatten the work into a standardized approximation of someone else’s voice. That is the line Herzog is drawing when he says the work becomes “so standardized” that the model can step in.
For developers, that line is worth keeping in mind when designing creative features. If a tool encourages users to chase resemblance too early, it may erode the very judgment that makes the result valuable.
The lesson is simple: a creative workflow should help people make choices, not just generate surfaces. Herzog’s comments are severe, but the underlying point is practical. When the process starts with imitation, the story often arrives second, if it arrives at all.
Top comments (0)