About us

What we hold to, and the product that follows from it

No origin story on this page, and no wall of customer logos — we do not have either yet. What we do have is a handful of positions already written into the code, expensive enough to reverse that they explain why the line works the way it does.

Who you are dealing with

The identity part, first

Operator: Octopus Core Pty Ltd
Company number: ACN 696 931 236
Business number: ABN 28 696 931 236
Registered in: New South Wales, Australia
Registered on: 8 April 2026
The product: An AI video production line
Built for: Vertical-drama studios and solo creators

The registered office is not printed here. Like the numbers above it, it sits on the public register held by the Australian Securities and Investments Commission (ASIC) and can be looked up from either the ACN or the ABN.

Six positions

Each one carries a “so”

“We believe in transparency” tells you nothing, because it rules nothing out. Every position below is followed by what the product does as a result — that second half is the part you can hold us to.

Direct vendor connections only

No aggregators like fal.ai or Replicate: they mark up the vendor's own price, take the margin, and add a layer of failure you cannot see into. No reselling finished products like Runway or Pika either — they are aggregating someone else already, so reselling them means paying two markups. And nothing sourced by turning consumer subscriptions into API access, which can be cut off unilaterally at any time.

Every vendor in the capability catalogue is a first-party account. Adding one is two methods, a catalogue entry and a set of contract tests — which is why swapping vendors is routine rather than an event.

We train no models and run no inference

Footage and speech come from external first-party vendors; screenwriting text goes through our own AI routing service. What we do is orchestration — arranging them into a line that delivers, and moving to the next vendor when one is not co-operating.

“Training our models on your content” cannot happen here, because there is nothing to train. What each vendor does with what it receives is governed by that vendor's terms, and the privacy policy says so plainly rather than promising on their behalf.

No model dropdown in the interface

Prompts, model names, seeds, image-to-video — those are internals of the production line, not homework for you. You pick a format, and the aspect ratio, runtime, shot density and narrative structure follow from it.

Zero-cognition is a product constraint, not an interface preference: no change is allowed to promote a model parameter into something you have to fill in.

Drafts are free, and so is falling back

The rounds you spend reworking a storyboard all run as drafts — structure, pacing, voice and subtitles are real, only the picture is a concept board, and nothing leaves the machine. When an external vendor fails, the line drops to the built-in implementation instead of stopping.

Drafts are the single largest saving on this line, and we hand it to you rather than charging per round. A degradation is ours, not yours, so every built-in implementation is priced at zero.

Masters carry an AI-generated badge by default

The badge is burned into the frame, not a boolean in the metadata. You can reword it for your own brand, and you can switch it off — but that is your compliance decision, not a default we make for you.

The default sits on the side of disclosure rather than the side of looking good. Flip a default like that once and nobody ever flips it back.

Model scoring reads our own line, nothing else

Four signals: reliability 45%, rework 25%, satisfaction 20%, speed 10%. All of them come from real production records — which vendor delivered, which shot got re-rendered, whether you rated a shot up or down, how long it took. Vendor leaderboards do not count here: different samples, different subject matter, different judging.

Too few samples means no score, rather than a number that merely looks authoritative. It is also why there is no leaderboard on this site — the public deployment has no production data behind it yet.

The other half

What we do not have

An about page made only of positions is an incomplete one. What follows is the current state of things rather than modesty — being willing to print it is a position too.

Not a video editorNo layers, no keyframes, no colour curves. For frame-level work this is the wrong tool.
No SLASingle-region deployment, no cross-region failover, no uptime commitment and no service credits.
No live vendor credentials in the public deployment yetA vendor without credentials is filtered out before selection, so what runs today is the built-in implementation.
No self-service account deletionRemoving an account and its data means writing to us for a person to action.
No customer counts, case studies or model leaderboardThey go up when they exist. Invented ones are the first thing to fall over.

Every claim on this page traces back to our code or to a written architecture decision. Where the trace is missing, the claim does not go up — and that rule is itself one of the six.

Better than reading this page: run an episode

Drafts are unlimited and cost no credits. Open an account, take one project the whole way through, and judge for yourself whether any of the above holds.

Start free — no credit card Read the AI disclosure