When people hear that I spent years working in feature-film visual effects, they often assume the most useful part of that experience was learning sophisticated software or technical techniques.
Those skills certainly mattered. Feature-film VFX requires patience, precision and a willingness to solve visual problems that may not have an obvious answer. Yet the longer I worked in the industry, the more I realised that the tools themselves were only one part of what I was learning.
The deeper lesson was how to think about an image within a production.
A shot was never judged only by whether it looked impressive on its own. It had to support the story, fit the surrounding sequence, meet the needs of the production and reach a standard appropriate for the audience that would eventually see it.
That way of thinking now shapes almost every decision I make in AI filmmaking.
A shot belongs to something larger
When you spend a long time working on a single shot, it is easy to experience it as a complete piece of work. You notice every edge, every movement and every small detail that could still be improved.
The audience rarely sees it that way.
They encounter the shot as one moment within a much larger experience. It follows another image, leads into the next and exists alongside dialogue, performance, music, editing and story. A detail that feels enormous while you are working on it may pass almost unnoticed in the finished sequence, while something that appears technically minor may completely disrupt the emotion of the scene.
Feature-film VFX taught me to keep moving between those two perspectives. The close view matters because the work must hold up, but the wider view matters because the shot ultimately exists for the film rather than for itself.
That has become especially important in AI filmmaking, where it is possible to become fascinated by individual generations. A beautiful image or impressive piece of movement can feel valuable simply because it is difficult to create.
The more useful question is whether it belongs in the film.
Technical quality is not the same as creative value
One of the most important lessons I learned in VFX was that a technically polished shot is not automatically a successful one.
A shot can be beautifully executed and still feel wrong for the sequence. The lighting may be impressive, the compositing may be seamless and the details may withstand close inspection, yet the image can still draw attention away from the story or create an emotional response that the scene never intended.
The reverse can also be true. A shot may contain small imperfections while doing exactly what the film needs. It communicates the right information, supports the performance and allows the audience to remain inside the scene.
This distinction matters because creative work often contains more opportunities for improvement than a production has time to pursue. Judgement is required to decide which changes will make the film stronger and which changes will merely make the shot more polished.
AI filmmaking presents the same challenge in a different form. A generated clip may be visually striking but unusable in context. Another may contain flaws while capturing the exact mood, composition or performance the sequence requires.
Evaluating those results demands more than recognising technical defects. It requires understanding what the image is supposed to achieve.
Imperfect does not always mean unusable
Feature-film production also taught me not to treat every visible problem as evidence that a shot should be abandoned, a mindset that also matters when deciding whether to regenerate AI content or build on what already exists.
A shot might contain an issue that could be addressed later, or a limitation that mattered less than the qualities already working within it. The important question was whether the foundation remained useful and whether the remaining problem genuinely prevented the shot from fulfilling its role.
That is a different mindset from expecting every stage to produce a finished result.
In AI filmmaking, the first reaction to an imperfect generation is often to create another one. Sometimes that is exactly the right decision, particularly when the composition, performance or underlying idea is fundamentally wrong.
At other times, regeneration replaces material that was already valuable. A later version may correct one flaw while losing the atmosphere, timing or emotional quality that made the earlier result worth keeping.
My VFX experience encourages me to pause before replacing something. I try to understand whether the problem invalidates the shot or whether the shot already contains enough value to justify moving it forward.
That pause is where production judgement begins.
Different stages solve different problems
A feature-film VFX pipeline contains many specialised stages because no single department or tool is expected to solve everything.
The work develops through a sequence of decisions. Different people address different problems, and a shot gradually moves towards completion through the combined contribution of artists, supervisors, production teams and filmmakers.
What mattered to me was not simply the number of stages. It was the recognition that each stage existed for a reason.
A problem discovered during one stage might be better addressed somewhere else. Trying to force every issue into the same part of the process could waste time, reduce flexibility or damage qualities that were already working.
This principle has strongly influenced how I approach AI filmmaking.
This connects closely to the Foundation idea that AI generation is only one step in filmmaking. Generation is powerful, but it does not need to resolve every production problem. Evaluation, editing, refinement and finishing each ask different questions. The exact boundaries between those stages may change as AI tools evolve, but the value of separating their purposes remains.
Once every issue is treated as a generation problem, regeneration becomes the entire workflow. That can make a project feel active while preventing it from developing into a coherent production.
Rough versions reveal expensive mistakes early
Feature-film production places great value on early versions because they allow important questions to be answered before the most expensive work begins.
Storyboards, previs, animatics and rough edits may look nothing like the final film, but they provide a place to test structure, pacing, staging and intention. Their value lies in what they reveal rather than how polished they appear.
This lesson continues to shape the way I work with AI.
Before spending too much time refining individual shots, I want to understand whether the sequence works. I want to see whether the story is clear, whether the pacing feels right and whether each image is earning its place in the edit.
A shot can be visually successful and still be wrong for the film. Discovering that after extensive refinement is far more costly than discovering it while the material is still rough.
AI changes how quickly these early versions can be created, and it may change which traditional stages remain necessary. It does not remove the need to evaluate the whole before polishing every part.
Production-ready does not mean nothing could be better
There were moments in VFX when everyone involved knew that a shot could still be improved.
Given more time, there might have been another detail to refine, another transition to soften or another creative possibility worth exploring. Yet productions cannot continue indefinitely, and every shot eventually reaches a point where it must move forward.
We sometimes referred to this as CBB, meaning Could Be Better.
The phrase was not an excuse for poor work. It acknowledged something that experienced artists already understood: creative work rarely reaches a state where nobody can imagine another improvement.
The real decision was whether those improvements remained important enough to justify delaying the production.
That idea has stayed with me because it separates perfection from readiness in a production context. A shot can still contain opportunities for improvement while already fulfilling its purpose for the film.
AI filmmaking often creates the illusion that one more generation is always available and that the next attempt may finally remove every compromise. In practice, the project still needs someone to decide when the work is ready to move forward.
The tools can create alternatives. They cannot make that decision for us.
Production constraints shape creative decisions
Working in a large production also taught me that creative decisions are never made in isolation.
A film has a schedule, a budget and many interdependent departments. A decision that seems small within one shot may affect work happening elsewhere, while an improvement that appears desirable may no longer be practical once the wider production is considered.
Constraints are often described as obstacles to creativity, but they also help clarify priorities. They force a production to decide what matters most and where additional effort will have the greatest effect.
AI changes some of those constraints. It can reduce the cost of experimentation, make certain forms of iteration faster and allow smaller teams to attempt work that once required far greater resources.
At the same time, it introduces new constraints. Generated results can be inconsistent, difficult to control and unpredictable across sequences. The abundance of possible versions can create its own form of uncertainty.
The role of production judgement is not to preserve old constraints simply because they are familiar. It is to understand which constraints have changed, which principles still matter and how the workflow should adapt.
Tools change faster than judgement
During my years in VFX, software changed constantly.
New tools appeared, established techniques evolved and tasks that once required complex manual work gradually became faster or more automated. Artists had to keep learning because the technical landscape never remained still for long.
AI filmmaking is moving even faster.
A workflow designed around the limitations of one generation of tools may become inefficient when those limitations disappear. Methods that once saved time may later create unnecessary work, while new capabilities can make entirely different approaches practical.
This is why I am cautious about treating any current AI workflow as permanent.
The procedure should evolve with the technology. The judgement behind it should remain focused on the film.
The most valuable lesson I carried from VFX was not how to preserve a particular technique. It was how to evaluate what a production needed when the available tools, time and constraints changed.
Filmmaking still comes first
AI makes it possible to create remarkable images, but filmmaking has never been defined only by the ability to create images.
It depends on choices about story, performance, composition, pacing, rhythm and audience experience. Technology can expand what is possible, but it does not decide which possibilities belong together or which ones are worth pursuing.
Feature-film VFX taught me to see technology as part of a larger production process. Even the most sophisticated visual effect exists to support something beyond itself.
That perspective is why I approach AI filmmaking as filmmaking first.
The generation may begin the image. It does not complete the creative responsibility.
Closing thought
My years in feature-film VFX gave me technical skills, but the most lasting influence was the production judgement that developed around them.
It taught me to evaluate images in context, preserve what is already working, solve problems at the appropriate stage and recognise when a shot is ready to move forward even though it could still be better.
Those lessons now shape how I work with AI.
The tools are new, but many of the most important questions are not.