Where Is AI Actually Useful in Creative Production?
We’ve been making things at Ghost Robot for more than 20 years, which means we’ve lived through a lot of tools that were supposedly going to change everything. Some did. Some didn’t. Most ended up somewhere in the middle: useful for certain things, useless for others, and eventually absorbed into the way we work until nobody talked about them much anymore. That’s roughly how we’re approaching AI. We use it. A lot, actually. We use AI for research, strategy, concept development, writing, visual exploration, pre-production, animation, post-production, and occasionally to create things that end up in the finished work.
But we don’t start projects by asking, “How can we use AI?” We start by asking what we’re trying to accomplish. Then we figure out which tools might help.
AI isn’t one thing
Part of the confusion around AI in creative production is that the term has become almost meaningless. “AI” can describe using a language model to summarize 200 pages of research. It can mean generating a reference image during concept development. It can mean removing an unwanted object from a shot, creating a synthetic voice, generating an entire video, extending a background, translating dialogue, analyzing an audience, or helping a producer sort through a mountain of information. Those are very different things. Asking whether we “use AI” is a little like asking whether we “use computers.”
Yes. The more useful question is where. Research is one of the least glamorous and most useful applications This is probably where we use AI most often. Creative projects tend to begin with a lot of information. Briefs, brand documents, audience research, meeting transcripts, competitive work, old campaigns, customer reviews, social conversations, product information, cultural references, and whatever else we can get our hands on. AI is very good at helping us navigate that material quickly. We can use it to organize research, identify recurring themes, compare large sets of information, surface contradictions, summarize interviews, cluster audience responses, or find a thread buried across several documents. That doesn’t mean we accept whatever it tells us.
Research still requires judgment. Sources need to be checked. Conclusions need to make sense. Sometimes the thing AI thinks is important isn’t important at all. But it can help us spend less time sorting and more time thinking. That’s valuable.
It’s useful for finding the question underneath the question
A brief might say:
“We need a video.”
Okay. Why? One of the most useful parts of strategy is figuring out what problem you’re actually solving before you start generating answers. AI can be helpful here because it allows us to interrogate a large amount of background material quickly. What is the audience saying? What assumptions does the brief make? What are competitors doing? Where are there contradictions? What questions haven’t been answered? Is there evidence that the proposed deliverable is even the right one? AI can’t make those strategic decisions for us. But it can help us explore the territory. And exploration is often where the useful stuff is hiding.
AI is an excellent creative sparring partner
We also use AI during concept development, although probably not in the way people imagine. We don’t type “give me 20 campaign ideas” into a chatbot, pick number seven, and call it a day. Generating ideas isn’t particularly difficult. Generating an idea worth making is. Where AI becomes more useful is in the back-and-forth. Push this idea further. What are we overlooking? What’s the obvious version of this? What’s the opposite? What assumptions are we making? How might this work as an event instead of a film? What would make this impossible? What happens if we remove the product entirely?
Sometimes the responses are terrible. That’s useful too. A bad suggestion can clarify why you don’t like something. It can send you in another direction. It can expose the cliché version of an idea so you know to avoid it. We’ve always believed that if you never have a bad idea, you probably aren’t trying hard enough. AI is extremely good at having bad ideas. Occasionally, one of them leads somewhere interesting.
How Integrate used AI before we made a frame
Our recent film for Integrate is a useful example of AI doing something more interesting than simply generating the finished work. Before Ghost Robot got involved, the Integrate team had spent months using AI to explore ways of expressing the company's vision. They weren't asking a tool to replace their point of view. They were using it to investigate possibilities, put language and imagery around abstract ideas, and get closer to what they wanted to say.
When we came in, we didn't just collect the outputs and turn them into a film. We spent hours talking with the team, including Brian Fiore, about the business, the people behind it, and what the brand needed to communicate. That human conversation mattered as much as the AI exploration. By the time we began making the film, we had a shared understanding of the ambition behind it and a creative direction we could actually execute.
That's a version of AI collaboration we find genuinely exciting. The tools helped the client explore and articulate a vision before production began. Our job was to bring creative judgment, brand voice, storytelling, and craft to that exploration, then make something distinctive out of it. AI didn't eliminate the need for a creative partner. It helped us start the partnership further along.
For marketing teams, there's a practical lesson here: use AI to bring more thinking into the room, not to declare the thinking finished. Bring the explorations, the strange experiments, the contradictions, and the half-formed ideas to your creative partner. Those can be much more valuable than a polished AI-generated treatment that nobody feels free to question.
Visual exploration has changed dramatically
This is one of the places where generative AI has had the most immediate impact on our process. Creative development used to involve a lot of searching for references that almost showed what we were imagining. The right lighting, wrong room. The right room, wrong person. The right composition, but everything else is completely wrong. You’d assemble a moodboard and spend half the presentation explaining which part of each image the client was supposed to be looking at. We still use references, photography, illustration, storyboards, design, and all the other tools we’ve always used. But now we can also generate images that get much closer to the thing we’re actually describing.
That’s incredibly useful during exploration. We can test visual directions before committing to them. Compare locations that don’t exist yet. Explore wardrobe, production design, composition, lenses, lighting, characters, environments, and visual styles. The point isn’t necessarily to create the final image. It’s to make the conversation more specific. “This, but warmer” is a much better creative conversation when everyone is looking at the same “this.”
Pre-production is really about making decisions
A surprising amount of production is just a long series of decisions. Which location? Which actor? Which wardrobe? Which prop? Which lens? Which set? Which shot? As we talked about in our piece about commercial budgets, more time and money usually gives you more opportunities to explore those choices before committing. AI can expand that exploration. We can visualize how an idea might look in a particular type of location before scouting it. Try different art-direction approaches before fabricating anything. Explore a prop before asking someone to build it. Create rough storyboards quickly enough to test several versions of a scene. None of those things eliminate the location scout, production designer, storyboard artist, stylist, or director. Ideally, they help those people make better decisions earlier.
That’s an important distinction.
Sometimes AI belongs in the finished work
There are also projects where generative AI is the production technique. That can make sense. Maybe the idea requires imagery that would be impossible or prohibitively expensive to create traditionally. Maybe the strange instability of generative imagery is part of the aesthetic. Maybe the concept specifically involves synthetic media. Maybe AI lets a small production create a world it otherwise couldn’t afford to build. [Integrate is a great example of this!] We’re interested in all of that. But the same rule applies that we’d use for stop motion, live action, CG, puppetry, animation, or any other production technique:
Why is this the right way to make this idea? “Because we can” isn’t usually enough. A technique becomes interesting when it serves the concept. AI can save money. But that’s not always the point. This is probably one of the biggest misconceptions we encounter. AI can absolutely reduce certain production costs. It can make some kinds of visual development faster. It can reduce repetitive work. It can create imagery that would otherwise require travel, fabrication, stock licensing, a shoot, or significant post-production. It can allow smaller teams to accomplish things that previously required larger ones. But that doesn’t mean an “AI production” is automatically cheap.
Good generative work can require enormous amounts of experimentation, iteration, compositing, retouching, editing, animation, sound, design, and human supervision. You might spend less money on a location and more money on an artist who spends two weeks getting a sequence to stop doing something bizarre with someone’s fingers. Sometimes AI removes costs. Sometimes it moves them. And sometimes the best use of the savings is not to reduce the budget at all, but to put those resources somewhere else where they’ll have more impact.
Speed is useful, but iteration may be more important
AI is often sold as a way to make things faster. It does that. But one of the things we find more interesting is that it can make iteration cheaper. If exploring five visual directions used to take several days, maybe now you can explore twenty. If creating a rough prototype was once too expensive to justify, maybe you can make three. If testing an unusual idea becomes easier, you might be more willing to test it. That can change the creative process. The danger, of course, is that you simply generate 200 versions and become completely incapable of making a decision.
More options aren’t useful unless somebody has taste. The value isn’t infinite iteration. It’s having enough freedom to explore and enough judgment to know when you’ve found something worth pursuing.
Where AI is less useful
There are also places where we think AI is frequently overused. Using it because it’s new isn’t a strategy. Using it to make generic images because they’re cheaper than hiring someone isn’t particularly interesting either. If the result looks like something anyone with the same prompt and software could have made, you’ve probably eliminated one of the things creative work is supposed to provide: distinction. We’re especially cautious when the value of the work comes from something real. A documentary about an actual person should probably involve spending time with that person. If you’re telling a story about craftsmanship, showing the craft might be more powerful than generating an approximation of it. If a performance depends on subtle human behavior, an actor may give you something you never would have thought to prompt.
Authenticity isn’t necessary for every project. But when it is, faking it more efficiently isn’t an improvement.
AI doesn’t eliminate craft
The tools keep getting easier to use. That’s good. But easier access to a tool doesn’t eliminate the need to know what good looks like. We’ve watched this happen before. Digital cameras made filmmaking more accessible. Editing software became cheaper. Visual effects moved onto laptops. High-end design tools became available to almost everyone. None of that made directors, editors, cinematographers, designers, animators, writers, or producers irrelevant. It changed what they could do. AI will do the same. The people we’re most interested in working with aren’t simply good at prompting. They’re artists and makers who understand composition, performance, storytelling, design, rhythm, light, movement, sound, and all the other things that made images interesting before generative AI existed.
Now they have another tool.
It doesn’t eliminate judgment either
This may be the bigger point. AI can give you more ideas, more images, more information, more variations, and more possible directions than any creative team could reasonably evaluate. Which means the scarce resource starts to become judgment. What should we make? Which idea is actually good? Does this feel like the brand? Is this true? Is this interesting? Has somebody already done it? Will anyone care? Should this even exist? Those questions don’t disappear because generating things gets easier. They become more important. So where is AI actually useful?
For us, the answer is fairly simple. We use AI when it helps us:
- Understand something faster
- Find patterns in complicated information
- Explore more possibilities
- Make an abstract idea easier to see
- Prototype before committing resources
- Remove repetitive work
- Solve a production problem
- Create something that would otherwise be difficult or impossible
Spend more of our time on the parts of the job that require human judgment We don’t need every project to use AI. We also don’t think ignoring it is particularly useful. After more than 20 years in a business where the tools, platforms, formats, and rules keep changing, we’ve become pretty comfortable with the idea that the way we make things is never going to stop evolving. The tools will change again. Our job is to stay curious about them without confusing the tool with the idea. Because the question that matters isn’t whether something was made with AI.
It’s whether it was worth making.
A decision you can make
Try AI-assisted visual exploration in preproduction, then evaluate the outputs against rights, consistency, and human review requirements before client use.
If you’re trying to figure out where AI belongs in a real creative workflow, bring us the problem before you pick the tool. We can help find where it actually makes the work better.
