The Problem Was Never the Models

Picture the room, because it is a real room and there are a few hundred thousand of them.

It is late. There are thirty browser tabs open. Eleven of them are model platforms, four are pricing pages, two are a forum thread from March where somebody claims one particular tool handles hands better, and the rest are documentation. There are seven active subscriptions, three of which nobody remembers signing up for. There is a credit balance in a currency that exists nowhere else on earth, and no way to work out what the next generation will cost until after it has cost it.

Somewhere underneath all of this, there is a film. It is a good film. It has been a good film for six weeks.

That room is where Pindow started.

What we were actually looking at

The founding instinct was simple and, we thought at the time, small. One of us had spent fifteen years behind a camera across advertising, features and documentary, the kind of career where you learn the grammar by being made to. And the observation was this: the AI work was taking longer than the real shoots had.

Not the generation. The generation was fast. Everything around the generation was slow.

So we did the unglamorous thing and counted. Over months, across our own productions and then across other people’s, we tracked where the hours actually went on an AI film. The result was consistent enough to be uncomfortable, and it broke into three costs that nobody was naming.

The search tax. There are now thousands of models. In any given week, somewhere between four and nine of them are genuinely worth using, and which nine changes constantly. A working filmmaker was spending entire days not making anything: reading release notes, running comparison tests, reading someone else’s comparison tests, and arriving at a decision that would be wrong again in five weeks. This is research labour, and it was being performed, unpaid and badly, by people whose job is to direct.

The switching tax. Even once you knew which tools to use, the work lived in pieces. Script in one place. Image generation in another. Video somewhere else. Upscaling somewhere else again. Sound, lip sync and edit accounted for three more. Every boundary between them was a place to lose a file, break a naming convention, drop a version, or re-establish a character who had been perfectly consistent twenty minutes earlier. The craft cost of this is worse than the time cost. Every context switch is a moment where you stop thinking about the film and start thinking about the software.

The clarity tax. This one made us angriest. Almost nowhere in this industry can a person answer a simple question before they commit: what is this shot going to take out of my budget? Credit systems obscure it. Tiering obscures it further. The same model can sit behind four different front doors with four different arrangements, and none of them tell you plainly. A director who cannot budget a sequence cannot plan it, and a person who cannot plan makes smaller work than they are capable of.

Three years of building came out of those three sentences.

The five problems we had to solve

One: which model, right now. This is the problem we spent longest on, and it is the reason the platform exists. The job is not to hand the user every model, because a list is the search tax with extra steps. The job is to know, for this shot, at this moment, with this intent, which model is currently best, and to route the work there without the user having to read a single release note. When the answer changes next month, it changes inside the platform rather than inside the user’s evening. The filmmaker’s decision should be what the shot is. Ours is what runs it.

Two: one surface. Everything on a canvas. Not a dashboard linking out to nine tools, and not a folder of exports, but an actual working surface where the script, the frames, the shots and the cut live in the same place and refer to each other. The reason this matters is not convenience. A film is a continuous object, and a pipeline that fragments it makes the filmmaker do the reassembly in their head.

Three: clarity you can act on. You should be able to see what a decision commits you to before you make it, in plain terms, in one place. This is not a technically difficult thing to build. It is a commercially inconvenient one, which is why so few people do it, and it is non-negotiable for us.

Four: consistency as infrastructure rather than effort. Characters, locations, style, grade. In most workflows these are re-established by the user, shot by shot, through sheer discipline and a folder of reference images. We made them objects the platform holds: cast once, carried everywhere. A film should not require you to re-cast your lead in every scene.

Five: tools that speak film. This is the part that only comes from having done the job. Shot size. Lens. Key light and its direction. Screen direction and the line. Eyeline. Coverage. These are not decorations on a prompt. They are the load-bearing vocabulary of cinema, and a platform built by people who have never stood on a set does not know which of them matter. The two hundred-plus presets and the prompt engine exist because somebody sat down and encoded a craft, not because somebody scraped a list of adjectives.

What that adds up to

Measured against the way we ourselves were working before we built it, the platform takes out up to half the waste in getting to a finished piece, and up to ninety percent of the time.

We would rather explain those figures than defend them, because a number without a mechanism behind it is just a poster.

The first saving comes from not generating the wrong thing. Routing a shot to the right model the first time, drafting rough before committing to the finish, and making the consequences of a decision visible before you take it. That is where the waste goes. Nobody is being clever. We are removing effort that should never have existed.

The time saving is the search tax and the switching tax, recovered. Those days spent comparing models, those hours spent moving files between platforms, those evenings re-establishing a character who should never have drifted. That work does not become faster. It stops existing.

Both figures are deliberately conservative, and we would like to be open about why. Every production is different. A thirty-second spot and a twelve-minute piece do not behave the same way, and someone who already has a disciplined workflow has less waste to recover than someone starting from thirty open tabs. We would rather quote a number we can stand behind on a bad week than a number that makes a better headline. In practice we tend to see more than this. We will let the work say so rather than the marketing.

And the quality does not move. We want to be clear about the shape of the claim, because this is the part that usually hides a compromise: the saving is not a trade. This is not a faster route to a slightly worse film. If a workflow saves you a week and costs you the shot, we have not built anything. We have built a quicker way to be disappointed. The only version worth shipping is the one where the finished piece is as good as the one you would have made the long way, and often better, because you still had the energy left to care about the last ten percent.

What we deliberately did not build

A short list, offered in the spirit of being judged on it.

• No upsell. No tier that exists to make the tier below it feel broken.
• No automation for the sake of automation. There is an agent that assembles a timeline from your shots, because assembly is bookkeeping. There is no agent that decides where the cut lands, because the cut is the film. We are aware that we could ship one and that it would demo beautifully. The captain’s chair is not available.
• No complexity as a signal of power. A platform that requires a tutorial before you can open it has failed before anybody generated anything. Restraint is not a limitation. It is what a tool looks like when somebody has done the hard work of deciding on your behalf.
• No pretending the model is the artist. The models are extraordinary, they are getting better monthly, and they will all be replaced. What sits above them does not get replaced: the grammar, the structure, the taste, the decision about what the audience should feel at second twenty-eight. That layer is what we built.

Who this is for

Both ends of the room, with the same console for each.

For someone opening a filmmaking tool for the first time, the promise is that they will not have to become an expert in model selection before they are allowed to tell a story. The best available tool is already pointed at their work. They get to spend their first evening on the film, which is the only reasonable thing to spend a first evening on.

For someone with twenty years of credits, the promise is different. Nothing is hidden from you, nothing is decided for you that you would want to decide yourself, and the parts of the process that were always a producer’s job are handled so that the parts that were always yours stay yours.

The four stages are the same for both. Script, where the writing and the catalogue of characters and locations live. Canvas, the infinite working surface holding the styles, the prompt engine and the generation. 3D Stage, where you block the shot spatially before a single frame is made, so the line, the eyeline and the key light are facts rather than hopes. Video Editor, where it becomes a cut.

That is the whole platform. It is deliberately not longer than that.

Why any of this matters

We should say the part that is not about features.

The thing standing between most people and the film they have in them has never been talent, and it has never been AI. It has been access: to equipment, to a crew, to a school that costs more than a family earns in three years, to the room where someone decides whether your idea gets a budget. The gatekeeping was never protecting quality. It was protecting gatekeepers.

What has changed is that the making is now genuinely within reach. What has not changed is that the reaching is still absurdly difficult for no good reason, buried under subscriptions and credit systems and thirty open tabs and the quiet, constant suspicion that you are using the wrong tool and everybody else knows something you do not.

Removing that is not a convenience play. It is the entire point. Every hour we take out of the process is an hour that goes back to the only part that was ever actually hard: deciding what the film is about, and then having the nerve to make it that.

Three years. One filmmaker’s irritation, a team of engineers who took it seriously, and a very large number of small decisions about what to leave out.

Think outside the window.