About

We build the boring half of AI.

The Logical AI is a small studio in Manchester. Most of what we do is unglamorous and invisible: the systems that sit behind a business and quietly do the work. Lead pipelines, automation, internal tools, content production. This guide came out of the last of those.

01What the studio does

We are not a content agency and we are not a consultancy. We build things that run.

PIPELINES

Lead systems

Sourcing, enrichment, outreach and follow-up wired together so it runs without anybody remembering to do it. The unglamorous plumbing that decides whether a funnel actually works.

AUTOMATION

Internal tooling

The dashboards, reporting and back-office jobs a business ends up doing by hand. Usually the highest-return AI work available, and the least talked about.

MEDIA

Content systems

Production at volume rather than one-off assets. Repeatable formats, written processes, and the quality control that stops output drifting.

METHOD

Written process

Everything we build gets documented as we go, because a system nobody can operate is not a system. That habit is the direct reason this guide exists.

Almost none of our work is visible. This is the one piece we decided to publish.

02Where this method came from

It started on the media side, as an internal question: could a genuinely difficult AI video format be made to work reliably, rather than occasionally?

The format chosen was first-person motion. A camera strapped to a body moving fast through a real place. Almost nobody was making it properly, which made it a decent test.

One account was built from nothing to find out.

10M+
Views across platforms in seven days
7
Days from launch
29
Failures isolated and fixed
£0
Spent on advertising

Getting there took several hundred generations, and most of them were unusable. Not slightly wrong. Unusable. Riders frozen mid-air while the background moved. Dives that came out as lazy diagonals no matter how the instruction was worded. Subjects that flew backwards, through walls, into buildings. Rejections with no explanation attached.

Every one of those cost real money and produced nothing.

After the third or fourth time the same problem appeared, it became obvious that these were not random. The same failures happen to everybody, in roughly the same order, whatever they are filming. They are behaviours of the underlying approach, not quirks of one video.

So they were written down, the way everything here gets written down. First as notes to stop repeating a mistake, then as exact wording that reliably prevented it. That internal document is what became the product.

03Why it reads the way it does

Most material about AI video is either a tool review or an inspirational thread. Neither survives contact with an actual generation that has gone wrong at eleven seconds.

A guide you cannot act on at midnight, with a failed generation open in another tab, is entertainment.
01

Symptom first

Every entry starts with what you are actually looking at on screen, not with a concept. You find your problem by recognising it, not by knowing its name.

02

The exact wording

Not the principle, the sentences. Paraphrasing them is how they stop working, so they are given verbatim and ready to paste.

03

The failures included

Rejected attempts are published alongside the finished ones, annotated with what was wrong. Nobody does this, which is why everyone repeats the same mistakes.

04

Honest where it is unresolved

Where something could not be worked out, it says so and gives the competing explanations. Confident guesses cost readers money.

04What we believe about this work

Position one

The model is not the advantage

It is the one part of this anybody can buy for the price of a subscription. Everyone has the same generator. What is scarce is knowing what to write into it, and that gap is where all the value sits.

Position two

The picture decides the video

Almost everyone spends their effort on the prompt and treats the opening image as a formality. It is the wrong way round, and rewriting words to fight a picture is the single most expensive habit in this field.

Position three

Realism comes from limitations, not from quality

Real cameras are limited. Generated images are not. Haze, bad light, overloaded microphones and blown exposure are what make footage believable, and every one of them has to be asked for deliberately.

Position four

Change one thing at a time

Change three and improve, and you have learned nothing. You will then carry all three forever because you are afraid to remove any of them. Discipline is cheaper than credits.

05Who is behind it

The studio is based in Manchester, United Kingdom, and is deliberately small. The work spans automation, lead systems and media production for businesses, and this guide is the media side of that written up properly.

The method comes from production, not from research. Everything in it was paid for in failed generations first.

06What we do not do

Worth stating plainly, so nobody wastes their time writing in.

  • We do not sell this as a service. There is no done-for-you video offer attached to this guide and no waiting list for one.
  • We do not sell software here. Nothing is installed and there is nothing to log into. It is a written document.
  • We do not run a community. No Discord, no cohort, no weekly calls.
  • We do not sell a subscription. One payment, yours permanently, updates included.
  • We do not promise results. The method is real and it is honestly reported. Nobody can guarantee a video will perform, and anyone who does is selling something else.

07Get in touch

Contact

Questions before buying, or something in the guide that contradicts what you are seeing, go to info@thelogicalai.com.

Corrections are genuinely welcome. Several entries exist because a reader wrote in.

See what is included