Interview by Mirco Sgarzi CCS
Artificial intelligence has become part of our work. Where did you first start experimenting with it? And in which areas do you find it most useful?
first started using AI mainly in the pre-production phase, for pre-visualisation. I used to spend hours on Photoshop creating images to show the director; now, with tools like MidJourney, I can generate striking visuals in a fraction of the time. This allows me to present visual ideas in a tangible way with much less effort. It saves a great deal of time and enables you to explore solutions you might not have considered. In fact, AI is not just a tool that executes commands—it sometimes produces images with unexpected details that make you think about new creative possibilities. Of course, it doesn’t always “understand” what you want; you need to learn how to “converse” with the machine, to use a “new language,” refine the dialogue, and give the right instructions. But once you master it, it becomes an incredibly powerful assistant.
The prompt was: “We are in a luxurious villa in Dubai, at a pool party with a DJ. A girl catches the eye, while guys and girls have fun—some drinking, some dancing, others swimming. The main colours are blue and pink, blending into the velvety atmosphere of the water, with some contrasting elements. Plenty of scattered lights in the background illuminate the scene, in a cinematic photorealistic style.”
As much as AI can facilitate certain aspects of our work, I wonder how important it is to maintain a balance. Don’t we risk losing creative control if we rely too much on these automated solutions? And if we all use the same tools, isn’t there a danger of aesthetic homogenisation?
Absolutely, it’s a delicate balance. AI can create a common ground, but it’s always the artist who makes the difference. The machine provides tools, but it’s up to us to use them in a unique way. The risk of homogenisation does exist, which is why it’s essential to preserve one’s creative vision and use AI as a support, not a replacement. Nowadays, AI can even create personalised profiles. Let me explain: if you and I enter the same prompt, the machine will generate two different results because, in my case, it applies a “personalised” profile based on what it has learned from my preferences. You can “train” it on your choices simply by making a binary 1/0 selection on a series of suggested images—similar to dating apps!
It seems unlikely that AI could truly replicate the intuition a cinematographer brings to their choices.
As much as AI can suggest interesting solutions, intuition and emotion remain essential. AI is a tool that assists us, but for now, it cannot replace decisions based on the feelings that develop on set, where the interaction between the director, actors, and crew creates a kind of magic that no machine can replicate. As long as there is a set, a director, and real-life actors, the human element will remain irreplaceable.
That’s exactly the point: what happens between people, in that exact moment, cannot be reproduced by any technology. However, I do acknowledge that AI allows us to explore new possibilities and speed up certain processes. For example, tell us how you used AI for the “throne room” location in The Promise series.
The director had a very specific vision in mind, but the location we had available was just a simple studio with four windows, previously used for another film. Using GANs, I generated a series of images that transformed that space into something grand and regal. I gave the AI precise instructions to create a hall with dim lighting, burning braziers, and scenic fires contrasting with the cold blue of the walls, with a beam of light backlighting the throne from above, leaving the king seated in shadow, creating a sense of mystery. These images were crucial for the director, who immediately embraced the vision I wanted to convey, and they also helped inspire the production design team regarding specific lighting elements I needed, such as flames and braziers.
-
-
-
-
The Promise – Throne room
Let’s dive into the technical details. You mentioned GANs. Could you explain what they are and how they work?
GANs are one of the most advanced types of artificial intelligence we use today. The acronym GAN stands for Generative Adversarial Network, which is a neural network that operates through a kind of competition between two sub-networks. The first network, the Generative Network, takes real images or synthetic data as input and generates a new output, trying to make it as realistic as possible. The second network, the Discriminator Network, checks whether what has been created by the first network is real or artificially generated, essentially acting as an opponent to the generator. This continuous competition mechanism allows for a dramatic improvement in the quality and realism of the produced images. Some artificially generated images are indistinguishable from real ones. Even with advanced testing (such as pattern repetition analysis or facial asymmetry detection) that I conducted using other AI models, it was impossible to determine whether they were synthetic or not. Incredible, isn’t it? Even now, I sometimes still believe those images are real and struggle to accept that they were artificially created!
Images created by Dallin Mackay with the Flux AI model
I understand! But let’s go back for a moment to a point that caught my attention. You mentioned a “new language” developing between humans and AI. Could you elaborate on this concept? It sounds fascinating.
When working with AI, you realise that it truly is a new language. We give instructions through “prompts”, textual descriptions that provide guidelines to the machine. At first, the results are often unpredictable or far from what we expected, but as we learn how AI responds to our inputs, the process becomes more refined. In essence, we are creating a shared language between humans and machines. Every new project is an opportunity to experiment, refine prompts, and understand how AI “thinks.” It’s an iterative process—with practice, we become more precise and achieve exactly what we want. An example? If I want to create a battle scene, I need to describe not just the visual elements, such as the landscape or the number of soldiers, but also details like shot composition, lighting style, and atmosphere. AI responds to these instructions and improves over time, just as the tools within the application evolve. It’s like training with a virtual assistant that becomes increasingly skilled at understanding our requests. As I mentioned earlier, a recent MidJourney update introduced the creation of a “personalised profile” to better understand user preferences. This is part of an ongoing improvement in the human-machine relationship, a “Cronenbergian” dynamic we will have to engage with.
It really seems that AI not only facilitates work but also helps communicate more clearly with other departments. You mentioned using AI to collaborate with the costume and production design teams. How did that go?
This is one of AI’s biggest advantages. For instance, in a recent scene, I needed to communicate a precise colour contrast between a character’s costume and the dark background. Instead of explaining it in words, I generated an image that visually demonstrated the exact effect I wanted to achieve. This kind of visual communication is immediate and far more effective. You can show the atmosphere, colour palettes, and even the lighting setup, ensuring that everyone works with the same vision. Production designers and costume designers can draw inspiration from these images to refine their creative choices, though of course, their final work remains the result of a much deeper human creative process.
Location scouting – reference image for AI
Prompt: “Reference image + a minister’s robe in dark blue, set in his palace, followers dressed in yellow and black, Tunisia, 1150 AD, cinematic visual style. Note: The request was different, but the AI generated a more interesting image”. Note: The request was different, but the AI generated a more interesting image than the initial idea.
This is a great example of how AI can facilitate collaboration, but I wonder—do you think AI will truly replace certain roles on set? Or will it always remain a technical aid?
We have no hope—AI will replace us, little by little. For now, AI can reduce work time and simplify certain technical steps, but I don’t believe it will completely replace key roles on set, as long as there is one. AI can enhance the pre-visualisation of a scene, create virtual environments, and even generate entire video sequences that closely resemble reality, but to me, cinema is about human relationships. That’s why I make films. What happens on set—the interaction between the director, actors, and the whole team (a family!)—is not something a machine can replicate. The set is a place where unpredictable moments occur, where an intangible magic is born through collaboration and the energy of people. AI will never replace that “erotic tension.” Come to think of it, documentary filmmaking could have a new future, precisely because it “documents” something real, instantaneous, and tangible—a countertrend. The first to disappear will likely be advertising, where planning and visual consistency between the agency’s request and the final result are already very close, while the emotions of the actors take a back seat to “technique.”
Are you exploring other AI applications? Can you share any updates on AI-powered video tools? Which AI-based tools are you already using in this field?
At the moment, I don’t use AI much to generate videos—the images still feel too synthetic, it almost scares me! But there are some truly interesting tools: Runway, Sora, Veo 2. Adobe has introduced AI features in Premiere, allowing you to extend a scene or fill in missing frames. AI generates the necessary frames while maintaining visual coherence. Even in the field of visual effects, I see huge potential. If you need to generate a background, add a detail, animate a 1920s photograph, or de-age a face (as we know!), AI can do all of this in seconds. Of course, it’s also an essential tool for VFX artists, as it can create and modify visual elements with a speed that would have been unimaginable just a few years ago.
Prompt: “A group of three young men, seen from behind, looks at the city of Tunisia in 1115 AD, in all its beauty and strength. The city is by the sea, with several ships in the harbour; the boundary between the forest and the city is clearly visible, marked by large walls and grand entrances. Distant point of view, cinematic, simple style.”
Do you think AI could create issues of “technological unemployment”?
Every innovation brings major transformations. People once thought painting was dead with the arrival of photography in the 19th century, or that digital technology would destroy film, yet it only created new roles—new professions like the Digital Image Technician. The same will happen with AI: some roles will change (many, in fact!), but new ones will emerge. Perhaps we’ll have virtual DOPs or “prompters”, as you called them! Professionals working alongside AI to create digital worlds. But I believe the essence of our work will remain, at least for this generation. The key is to adapt and keep developing our skills. Another crucial topic emerges: copyright. Who owns AI-generated images? The person who wrote the prompt? The machine? Who owns an actor’s digital face? There will come a time when these issues become critical. This is the greatest revolution of the last century, and with it comes a host of complex ethical questions. Perhaps “knowledge” itself will no longer hold the same value as it does today. Let me explain: if a machine possesses all knowledge, what will be the point of consulting a lawyer? … Quite unsettling
Could you explain how to craft an effective prompt in MidJourney? What are the key principles behind it?
A prompt in MidJourney is a text string that describes the image you want to generate. The fundamental rule is that the most important elements should be placed at the beginning of the prompt. The AI assigns decreasing priority to elements as they are listed. A prompt can include details such as the main subject, style, perspective, and even visual references, like a painting style or that of a well-known photographer. For example, if you want to create an image based on a photographic reference, you would first upload the image (photo 1 pig), then describe it in order of importance: “Dark and foggy night, a dystopian future where animals roam an Italian city like Bologna, two girls walking in the background, soft colours, realistic style, vintage look — ar 2:1 (which defines the aspect ratio).” This way, you’ll generate an image containing the specified elements, and you can adjust how closely it adheres to the reference or how much it diverges, using predefined commands. MidJourney has become more sophisticated, introducing new tools to manage element priority and stylistic consistency more effectively. It may seem complex at first, but with practice, you learn the correct language to communicate with the AI and achieve your vision. Though, more often than not, AI surprises you with solutions you never anticipated.
Will there come a time when AI will be indistinguishable from reality? What do you think will happen to our perception of the real?
That time has already arrived—at least for still images. And by the time people read this article, perhaps even for video… This is both the most unsettling and the most fascinating aspect. Today, AI is a tool that supports us, allowing us to create increasingly realistic images. But there will come a day, not too far off, when AI-generated fiction will be so perfect that it will no longer be distinguishable from reality. When we reach that point, we will be forced to completely rethink the very concept of “real.” It will be a world where we can no longer say with certainty whether what we see on screen is real or artificially created. This raises profound questions about what “reality” means and how we will interpret the world around us. Will we still be able to trust our perceptions? Or will we constantly have to question what is true and what is not? This boundary between reality and fiction will become increasingly blurred—perhaps to the point where there is no clear distinction anymore.
Thank you, Alessandro.
