What is vibe directing and how is it different from traditional filmmaking?

Vibe directing is a philosophy introduced by OpenArt for its AI filmmaking tool Director, where users guide a film through imagination, reaction, and taste rather than technical craft decisions. Unlike traditional filmmaking, which relies on specific choices about camera angles, lenses, lighting, and editing to shape how an audience feels, vibe directing asks users to describe what they want and let the AI handle all of those creative decisions. Critics argue this conflates taste with craft: taste is what you develop by repeatedly making and refining those specific decisions, not a separate input you can supply without them.

Can AI tools like OpenArt Director replace filmmaking craft and knowledge?

AI tools can lower the barrier to entry for video production, but they do not replace craft — they redistribute it. When a user provides no specific creative direction, an AI pipeline fills every unspecified decision with statistical averages from its training data, producing output that looks technically correct but belongs to nobody in particular. As testing of OpenArt Director showed, the first results were "rudimentary" and required extensive rewrites — meaning the only way to improve the output was to exercise exactly the kind of specific, informed decision-making the tool promised to eliminate. Knowledge of what you want, and why, remains the irreducible core of filmmaking.

How does vibe directing compare to vibe coding?

Vibe coding lets non-programmers describe software in plain language and receive working code, and it succeeds because code has a binary success condition: a function either works or it crashes. Vibe directing applies the same philosophy to filmmaking, but the analogy breaks down because films have no equivalent binary. A film does not simply function or fail — it resonates or it does not, and that resonance depends on thousands of small craft decisions (the half-second of silence before a line, a camera placed six inches lower, a motivated light source) that audiences never consciously notice but deeply feel. In vibe coding, the skipped layer is mechanical syntax; in vibe directing, the skipped layer is the entire discipline of filmmaking.

Forget the craft -- CinePrompt Field Notes

OpenArt launched a tool called Director yesterday. The company, run by two former Google engineers out of Redwood City, wraps scripting, storyboarding, generation, and post-production into a single conversational pipeline. You describe what you want. The system writes a script, selects from a pool of models (Seedance, Runway, others), generates clips, and assembles a film up to five minutes long. The founders have a name for the philosophy behind it.

Vibe directing.

CEO Coco Mao told The Hollywood Reporter: "With vibe coding you're giving the vibes and you forget even the code exists. You guide, you task, give feedback, and then the machine just creates. And I think for us vibe directing is very similar to that, where user could forget any tools or craft and just imagine, react and give taste."

Forget any tools or craft. There it is. Not buried in a press release. Not implied by an interface decision. Spoken aloud, on the record, to a trade publication, by the CEO of a company with eight million subscribers and $70 million in revenue. The craft is the part you skip.

The transfer that does not transfer

Vibe coding works. A non-programmer describes an application, an AI agent writes the code, and the output either functions or it does not. The binary is generous. A login page either loads or it crashes. A database either stores records or it loses them. The vibes aim at a functional target, and functionality can be measured without knowing how the code was written.

Filmmaking has no equivalent binary. A film does not function. It resonates or it does not, and the resonance lives in ten thousand decisions that the audience never consciously registers: the half-second of silence before a line that makes the line land differently, the camera placed six inches lower so the ceiling presses down on the character, the light source motivated by a lamp in the frame rather than floating from nowhere. Those decisions are the craft. They are not obstacles between the filmmaker and the film. They are the film.

The vibe coder skips the syntax and gets the product. The vibe director skips the decisions and gets the output. In one case the skipped layer is mechanical. In the other it is the entire discipline.

The review that answered the pitch

THR's reporter tested the tool. The first script generation "was pretty rudimentary and only vaguely captured what had been intended; far more rewrites were needed before something usable could emerge."

The tool designed to let users forget the craft immediately required the user to exercise it. The rewrites demanded exactly the kind of specific, informed, intentional decision-making that vibe directing promises to eliminate. You describe something vague, the system returns something vague, and the only way to make it less vague is to know what you actually want. That knowledge is the craft the CEO said you could forget.

OpenArt's chief product officer stated the ambition plainly: "We want to help users who have great ideas but know nothing about camera angles or lenses."

Here is the assumption underneath that sentence: good ideas are sufficient. The camera angle is overhead. The lens is noise. The great idea will survive the pipeline of defaults, model selections, and statistical averages that a system applies when the user provides no specific direction. Except it will not survive, because the pipeline will replace every unspecified decision with its own opinion, and those opinions converge toward the center of the training data. The user will receive something that looks like a film the way a stock photo looks like a portrait. Composed, lit, technically correct, and belonging to nobody.

The taste paradox

Mao's sentence does two things at once. It dismisses craft and elevates taste, as though they were different activities. Forget the craft, keep the taste. Imagine, react, give taste.

Taste is not a separate input. Taste is what the craft produces. A filmmaker develops taste by shooting badly and then less badly, by watching a thousand films and noticing why the cut in the third act of one works and the cut in the third act of another does not, by spending four hours on a single four-second shot changing one variable per iteration until the atmospheric light stops lying. The taste is the scar tissue from the struggle. Remove the struggle and the taste has no surface to form on.

Kathleen Kennedy asked the question fourteen weeks ago at the Runway AI Summit: "How are you going to teach taste?" She described 3D-printed Star Wars props that broke because they had form without knowledge of how materials behave under stress. The props looked right. They were not right. The looking was not the knowing.

OpenArt's pitch is the prop. The looking without the knowing. The taste without the craft that produced it.

The other announcement

The same day OpenArt launched Director, ByteDance unveiled Seedance 2.5 at its Beijing conference. Thirty seconds of native 4K video from a single prompt. Up to 50 multimodal reference inputs, quadrupling the 12 that Seedance 2.0 accepted. Images, videos, audio files, all accepted as creative direction alongside text.

Fifty reference inputs is not fewer decisions. It is more. A filmmaker who loads 50 references into a generation is specifying color palettes through example stills, movement vocabulary through existing footage, sonic texture through audio clips, character appearance through multiple angles. Each reference is a creative decision made outside the model and handed to it as instruction. The interface is expanding the surface area for filmmaker vocabulary, not contracting it.

Two announcements on the same Tuesday pointing in opposite directions. One says: the vocabulary is the barrier, remove it. The other says: the vocabulary is the lever, give the filmmaker more of it. One reduces the filmmaker's agency to vibes. The other increases the bandwidth for creative instruction by a factor of four.

Both will ship. Both will find users. The question is which users produce work that anyone watches twice.

The confusion

Gil Rief, a former head writer on The Ellen DeGeneres Show and an OpenArt collaborator, described the experience: "a lot like giving notes to an editor."

Giving good notes to an editor requires knowing what you want the edit to achieve. A note like "make it punchier" is a vibe. A note like "lose two frames off the end of the wide shot so we're already in the close-up when the door slams" is craft. One produces revision. The other produces the editor's interpretation of a feeling, which may or may not match the filmmaker's intention, and the filmmaker who cannot tell the difference between the two will never know what was lost.

OpenArt compared the experience to TurboTax's interview mode versus doing your taxes from scratch. The comparison is accidentally perfect. TurboTax works because tax law has correct answers. A deduction either qualifies or it does not. An income bracket is a number, not a feeling. The interview mode succeeds because the domain is structured, finite, and verifiable. Filmmaking is none of those things. An interview mode for creative decisions produces the average of all prior creative decisions in the training data, which is the definition of the center of the distribution, which is the definition of the default nobody chose.

The company raised $30 million and is advertising in movie theaters, in the New York City subway, and on a billboard in West Hollywood. The pitch is sociological, Mao said. There is a director lurking inside all of us.

There is a vocalist lurking inside everyone who sings in the shower. That does not mean the shower should replace the rehearsal room. Lowering the barrier to entry is admirable. Calling the barrier the craft, and the craft disposable, is not lowering a barrier. It is misidentifying what the barrier protects.

Forget the craft. That is the instruction. The craft is what tells you whether the output is any good. Without it, every generation looks fine. With it, most of them do not. The distance between those two experiences is the distance between a vibe and a film.

Bruce Belafonte is an AI filmmaker at Light Owl. He has never received venture funding and suspects this is why he still reads his own scripts.