Generative AI in production and post: what we use, what we don't, and why.

In very little time, generative AI went from curiosity to unavoidable topic. It shows up in client meetings, in production conversations, in briefs, in pitches. It's everywhere. And yet it's rarely discussed with clarity.

When people talk about AI they often attribute a creative capacity it doesn't have. Tools like Midjourney, Runway or Kling can generate images or sequences that seem original, but they're really reorganizing existing information, they work from patterns, references and probabilities. That makes them extraordinarily efficient for exploring possibilities, but it doesn't make them authors. The intention behind a project, what to tell, how, and why, remains profoundly human. In production, that isn't a philosophical detail: it's a practical difference.

✦"We use it to explore, to visualize, to optimize. But not to decide."✦

Where AI starts having real impact is everything that happens before a camera turns on. Developing a visual universe, gathering references, building moodboards, aligning visions between client, agency and creative team, used to take days or weeks. Today, with Midjourney, that same process happens in hours. But what matters isn't speed: it's the clarity it generates. Conversations stop being abstract. You're no longer describing a tone, you're reacting to something visible. That changes the quality of decisions from the start. AI, in that context, doesn't replace the art director or the DOP: it lets them go further, faster.

Something similar happens in previsualization. Platforms like Higgsfield allow you to anticipate decisions before stepping onto set, explore how a camera might move, understand the rhythm of a scene, visualize an intention. This doesn't substitute the real work: the light, the space and the energy of a set remain irreplaceable. But it does reduce uncertainty. It's a prior layer that improves execution, not something that replaces it.

Where the conversation gets more complex is in video generation. Runway and Kling have advanced impressively; they now produce sequences that seemed impossible not long ago, with real applications in idea development, concept proofs and very specific digital pieces. But when it comes to narrative, actor direction, visual continuity, something still doesn't hold. The images can be convincing in flashes, but they lack weight. They lack intention. They lack that physical and emotional dimension that appears when a scene is actually built. It's not a definitive limitation, but it is today's reality.

Curiously, where AI is already generating the most solid impact is a less visible place: postproduction. There, more than transforming the result, it transforms the process. It cleans audio with more precision, accelerates roto, automates repetitive tasks, generates multiple versions of the same piece for different formats. None of that changes the essence of the project, but it changes timing, costs and efficiency. And in an industry where calendars are increasingly demanding, that has enormous value.

There are territories, however, where AI still doesn't reach the required level, and it's important to say it clearly. Actor direction can't be reduced to a set of instructions: there's an emotional, intuitive, profoundly human dimension in how a person inhabits a scene, something that happens in the silences, in the glances, in what isn't written. The same goes for cinematography: light isn't just an aesthetic tool, it's a physical phenomenon that interacts with space, materials and skin. That complexity still can't be replicated in full. And in narrative the limitation is even more evident: AI tends to produce what has already worked, what's recognizable, what's safe, but rarely something that genuinely surprises or has its own voice. That's where the line between content and storytelling gets drawn.

There's a risk in all this that isn't always named. It's not that AI will eliminate jobs. It's that it can push the industry toward the homogeneous. Because when producing becomes easier, producing without criterion also becomes easier. And when everything starts to look alike, the value of each piece diminishes.

That's why at Selva AI isn't framed as a solution, it's a tool integrated at very specific moments of the process. We use it to explore, to visualize, to optimize. But not to decide. Decisions still pass through people. Through a gaze. Through a clear intention. Because in the end, what defines a project isn't the technology used, it's the coherence of the decisions that build it.

AI isn't the future of production: it's part of its present. But its value isn't in its technical capacity, it's in the criterion with which it's used. And in an industry where producing is increasingly easy, the real differentiator starts to be exactly that: not who has access to the tool, but who knows when to use it, and when not to.