Your Screenplay Isn’t Too Vague for a Director, but It Is for an AI Model

The page from Lost Garden, an AI anime series, read well on paper. A single line described a heroine torn between trusting a stranger and fleeing. Any script reader would have understood instantly. However, an AI model transformed it into a woman standing still, displaying an expression that could signify boredom or mild indigestion.

The problem wasn't the line itself, but rather it had been written for the wrong reader. Screenplay pages have traditionally fulfilled two roles: describing what is on-screen and trusting the reader to visualize it. For a century, that reader was a director, a director of photography, and an actor – individuals with a wealth of context to fill in the blanks left by the page.

AI video models, on the other hand, lack this context. They lack the ability to infer or translate. If a line describes a feeling rather than a specific frame, the model must invent a frame from scratch, often resulting in an inaccurate interpretation. This isn't a plea for a new kind of screenplay. The same rule should apply, just without the human reader's assistance.

The crux of the issue is who will read the page. Action lines, regardless of the format, remain the same. The change lies in the recipient of the page. "Show, don't tell" has always been the foundational principle of screenwriting, even before the advent of AI generation. No Film School's analysis of action line craft and Backstage's guide on writing action lines both reiterate this standard: action lines should only describe what the audience can see on-screen, written in present tense, and without any visually unfilmable elements.

This rule was always intended for a human crew who could still read between the lines. However, now, the words are the only thing left. A human assistant would interpret "she hesitates, torn" as a held beat, a glance at the door, and a hand that almost reaches out. An AI model, however, has no equivalent move to make. It doesn't know what "torn" looks like on a face it has never directed.

So, it generates something close to average, resulting in a shot that is technically compliant but dramatically empty. The emotion was never actually captured in the shot. It was always something a human inferred for them. The guiding principle "show, don't tell" holds more significance now than ever before. Professional format guidance emphasizes that if it isn't visual, it isn't necessary in an action line.

This isn't a new AI-era rule, but rather a tightened version of a rule that good scripts have followed for decades. A few habits that were once forgivable no longer are. Interior states devoid of visible equivalents, adjectives describing a category rather than an image, and backstory folded into the action line all fall under this category.

Passive, vague verbs also pose a challenge. "The door is opened" leaves the who and how ambiguous. "She kicks the door open" does not. Finally, a few tips on rewriting interior beats into visible ones. The line that initially failed on Lost Garden was "she is torn between trusting the stranger and running." The solution wasn't to add more description, but to identify the one physical action that embodies the emotion and write only that.

Before: She is torn between trusting the stranger and running. After: She takes one step toward the door, stops, looks back at him over her shoulder. This revised version is shorter and accomplishes more. A director would have made this translation silently in their head and never shared it. An AI model cannot, so the translation must be made on the page before generation.

This is essentially a two-step habit: ask what the audience would actually see, not what the character is feeling. If you can't answer with a single physical, present-tense image, the line isn't complete. Front-loading specific details that the shot depends on also helps. Whether it's light, distance, or a single gesture carrying the beat, these elements should be placed at the beginning of the sentence, rather than buried in a series of clauses.

Implementing these changes consistently results in a page that is more filmable for human crews as well, rather than just AI models. Action lines, despite the format change, still adhere to the same craft standards. AI has merely removed the option to skip this step. The action line ends where the shot list begins. They are not the same document, and conflating them can lead to confusion.

The action line is prose, written to be read and intended to engage financiers or readers before a single shot is created. A shot list, however, is a technical breakdown, detailing duration, camera angles, continuity anchors, and other elements consumed by the generation pipeline. A disciplined action line does not replace a shot list; rather, it aids in its creation by allowing half of the visual decision-making to be completed on the page instead of being passed on to the breakdown stage.

This is the reason behind the development of ScreenWeaver, a tool designed to keep the script, visual reference, and shot notes interconnected instead of becoming separate documents that drift apart and create inconsistencies.