Skip to main content

The annual

Video generators resist comparison.

Comparing video generators is harder than comparing almost anything else in software, and most published comparisons quietly skip the part that makes it hard. This is an annual that intends to say what it actually knows.

The difficulty

Why these tools resist comparison

Four things, each of which would be enough on its own to make a league table dishonest.

The same prompt gives a different film every time

Generation is stochastic. Run one prompt five times and you get five clips of varying quality. Any single output is therefore a sample, and a comparison built on one generation per tool is comparing luck as much as capability.

Demo reels are selected, not representative

A showcase is the best of many attempts, chosen by the people who made the model. That is not dishonest of them — it is a showreel. It stops being reasonable when someone else presents it as evidence of a typical result.

Prompts do not transfer between models

Each model responds to different phrasing, and a prompt refined against one will usually underperform on another. A “fair test” using identical prompts is fair in form and systematically favours whichever model the prompt was written against.

The units are not the same

Clip length, frame rate and output resolution vary widely, and holding quality steady gets easier as clips get shorter. Comparing a four-second clip against a ten-second one tells you very little about either.

What is worth comparing instead

The things that hold still

Output quality moves with every model release. These do not, and they are usually what decides whether a tool fits a job.

What you can put in

Text only, or an image you supply, or a first and last frame, or a motion path drawn over a still. This changes what the tool is for more than any quality difference does.

How long a clip can be

A published, checkable number, and the one that decides whether a finished piece is a single generation or a stitching job.

What the licence permits

Commercial use, client work and resale are treated differently by different vendors, and free tiers frequently exclude all three. This is written down and rarely read.

What a generation costs

Credits per second of output, whether failed generations are refunded, and whether credits expire. Also published, also checkable, also the thing people discover late.

Whether a result is reproducible

Some tools expose a seed so a good generation can be recovered or varied deliberately. Without one, a result you liked is gone.

Where the output ends up

Resolution, frame rate, file format and whether a watermark is applied. All of it decides whether the clip can be used where you intended it to go.

House rules

How this will be written

No invented numbers

No scores out of ten, no star ratings, no tier boards. Where something is better, it gets said in words and the reason is shown.

Only claims we can stand behind

Nothing describes a test that did not happen. Pricing and feature claims are checked against the vendor's own pages.

No paid placements

Nothing here is sponsored, and outbound links carry no affiliate tracking and earn no commission. No company pays to be included or to be written about more kindly.

Status

This site is new and there are no comparisons on it yet. Despite the name, nothing here will ever hand out a placing or a medal — an annual can be a considered write-up without pretending to have measured something it has not. When comparisons land they will say how many generations they are based on.