# ROUTIX Blind

> ROUTIX's own perception and reasoning core, built to rebuild a scene the way a mind that once could see does — behind the chat and behind Optic.

**Every video ROUTIX touches is seen first by ROUTIX Blind.**

ROUTIX Blind is ROUTIX's own perception and reasoning core — designed and built by the ROUTIX team. You never open it and you never pick it. It works behind the agent in the chat and behind Optic, and every decision they make starts from it: what is in your video, who is who, what you are asking for, what must change, and what must not.

It takes in video, pictures, sound and words together. It builds a layered picture of every shot. It remembers in two ways — in pictures and in words. It carries every character, place and thing across your film. And it reasons again and again before it acts.

It is called Blind because of how it sees.

## A mind that once could see

Imagine someone who could see, and then lost their sight. They have not lost the world. They still know what red is — how it feels, where they last saw it. They know how late-afternoon light falls through a window, how a shadow stretches across a floor, how cloth folds, how a face looks a moment before it smiles. Perspective, depth, texture, the weight of things in a frame: they know all of it. They just do not see now.

Someone blind from birth has never known red. Someone who once could see keeps it for life. ROUTIX Blind is built on that difference.

Describe a scene to that mind and it does not paint a flat picture. It **rebuilds** the scene: the place first, then the light, then who is there, then every detail and where it sits, then what moves and when. And it does not stop at a still — it builds a film. Tell it that a man at a café table lifts his head and smiles, and it fills in every moment between: the head rising, the shadow under the chin changing shape rather than vanishing, the cup still on the table, the coat still black.

If, halfway through, the light suddenly came from the other side, or the cup disappeared, or the coat turned grey, that mind would stop. Its picture would break. That is exactly what goes wrong in video made by AI — and exactly what ROUTIX Blind is designed never to allow.

There is a reason every character in ROUTIX's branding is blind. They stand for the way ROUTIX thinks.

## Why the Blind mindset

ROUTIX did not choose this way of thinking on paper. In our tests we tried many paths and many workflows, and along the way we rethought rules most tools still follow. What held up, again and again, was the mind on this page: a mind that once could see.

Why that mind? Because a revision is only as good as the understanding behind it. Look at a video the way an eye does, and you see surfaces — and every change you ask for has to argue with them. Rebuild the scene the way that mind does, and you know what each thing is, where it stands, how the light reaches it and what depends on what. Then one thing can change, and the rest stays whole.

We will not ask you to take our word for any of it. We would rather talk about the result you get. Bring a clip, revise it, and look — the quality is in front of you, and the judgement is yours. Measure it against any revision model there is today, or any that comes after.

## The research

ROUTIX Blind did not come from a name. It was developed over two years of research and engineering, with more than 500 people who lost their sight after having seen — studying how they imagine, how they remember, and how they rebuild a scene in the mind from nothing but words.

Before any part of it was built, the ROUTIX team ran 222 experiments and studies, to learn what to build. Around them: thousands of hours of conversation, trial and error. The experiments took place at ROUTIX's own place, with the full consent of everyone who took part. What they taught us became the way ROUTIX Blind thinks — and they shaped Optic too.

Most of that work we are not sharing yet. Here are three of the experiments.

### One describer, five listeners

A sighted person sits facing five participants, each of whom once could see, and describes a painting or a picture to them. Each participant is asked to narrate their imagining aloud — not what they think the picture is, but how it forms in them, from the very first moment of thought to the last.

For example, the describer begins: *"A wooden table by a window. Late afternoon. The light comes in low from the left and lies across the table…"*

For example, our first participant: *"The light comes before anything. I put the window on my left and let it pour in, low and warm — and everything else waits for it. Now the table can come, half in that light."*

For example, the second: *"I start with the wood. I know its grain before I know its colour. Then you say late afternoon, and it turns to honey where the light touches it and goes almost black at the far edge."*

For example, the third: *"I need the room first. A window means a wall, and a wall means corners. I build the space around that window and keep the corners dark — then the table has somewhere to stand."*

For example, the fourth: *"I see the air. Dust in the beam, slowly turning. The table is under it, and the dust tells me how strong the light is."*

For example, the fifth: *"I put myself in the doorway first. Now I know where I'm standing, the window is ahead and to the left, and the table is in front of me, its near edge closer than its far one."*

Five paths — through the light, the material, the space, the air, the viewpoint. Five different ways of reasoning. And each of them arrived at the picture, vivid and whole. It held even when the describer was describing something that existed only in their own imagination — a picture nobody had ever painted.

The lesson we carried into ROUTIX Blind: the path to a scene can follow whatever you say first, and the scene still has to come out whole — place, light, things, the space between them, and time.

### Two minds, one picture

Two participants, each of whom once could see. The first is given a picture, or an event, in full detail. In words alone, they pass their inner picture to the second — until the second holds the very same image.

For example, our first participant begins: *"A narrow street just after rain. Evening. The lamps are already lit, and every one of them is doubled in the wet stones."*

For example, the second: *"Which side are the lamps on?"*

*"The left. And the street bends to the right, away from us."*

*"Then the bend is darker. The last lamp lights the corner of a wall… and the stones there have no reflection, because no lamp reaches them."*

*"Yes. And there is one window lit above the bend."*

*"One window. Yellow. Small. It's in the reflection too — broken by the stones."*

*"That's it. That's the street."*

Each of them reached the image by their own route — and it was the same image. Nothing was lost between them, because both minds know colour, shape, light, physics and texture, and neither had anything in front of their eyes arguing back.

### Imagination under emotion

How does a mind imagine when it is angry, or joyful, or stirred up? When a description reaches someone in that state, how fast do they build the picture — what do they keep, what do they drop, and when does the picture still come out whole? This experiment asked exactly that.

For example, the describer: *"A market at noon. A red awning over a fruit stall. Crowds. The sun straight overhead."*

For example, a participant who is angry: *"Market. Red awning. Noon — so the shadows are short, right under everything. Fine. I have it."*

For example, a participant who is joyful: *"Noon in a market — I can hear it. The awning glows red, the fruit underneath is half in its shade…"*

One picture built fast and spare, one slow and full — and in both, the place, the light and the red awning, where they belong.

The lesson we carried into ROUTIX Blind: hold the whole scene and its essentials even when the request is brief, or hurried — so that a few words are enough to change one thing without losing anything else.

*These exchanges are illustrations; our participants' own words stay private.*

The research ran far wider than these three, and much of it stays with us for now. We plan to share more of it in time.

## The insight: a tree, and a car

Point at a tree in a photograph and tell a sighted person, *"This is a car."* Their eyes argue back: *that's a tree.* However hard they listen, they drift back to their own reading of the picture. The pixels win.

Tell the same thing to a mind that once could see, and it simply builds what it was told: a car, standing where the tree stood, in the same light, its shadow falling the same way, its paint taking the same afternoon glow. No argument. No *"I don't see it, so I don't believe it."* A mind that knows colour, shape, physics, texture and light — free of whatever is in front of its eyes — builds exactly what is described, whole and coherent.

That is the heart of ROUTIX Blind. When you ask for a change, it does not argue with your old pixels. It rebuilds the scene from understanding — your scene, with your change inside it — and imagines it whole. That is why Optic can turn what is there into what you describe, and keep the world around it as it was.

## Why "next generation"

We call ROUTIX the next-generation revision platform for one reason. What stands behind Optic's revision reasoning is not one model, and not one workflow. It is a way of thinking: the mind of someone who once could see — with an imagination that does not stop until the scene is whole.

Take a replacement. Someone in your clip is to become someone else — taller, perhaps, or smaller. An eye that is looking at the picture starts from what is there: it lays the new person over the old one, the new height over the old height, and patches whatever does not fit.

A mind that once could see works the other way round. For it, the old person is gone — there is nothing left to lay anything over. It rebuilds the space first: the room, where the light comes from, the colour of the sky outside. Then it brings the new character in, with care for everything around them: how that light falls on them; what they change around them; what depends on them — what they wear, what they hold, what they touch; and, when they are not the old one's size, what that does to the people and things around them, and what those, in turn, do to the others.

That is the way of thinking ROUTIX's revision reasoning is built on. In Optic, it already stands as firm rules:

- **The whole stretch is read at once,** before anything is made — every shot, every subject, and one identity for each across the cuts.
- **What depends on what you change is followed.** What it wears, layer by layer; what it carries or holds; the marks on it; what goes with it; its state — each found and followed moment by moment, so it changes with your change, or stays as you say.
- **The new one plays the old one's part.** It stands where the old one stood, at the same size, in the same pose and on the same path, and plays your clip's performance frame for frame — the eyes, the lips, every expression — whether or not anyone speaks.
- **What happens to the old one happens to the new one,** at the same moments and only where your clip shows it — a mark, damage, a glow.
- **The light that falls on it is your clip's** — the same direction, strength, shadow and colour of light on every part of it.
- **The world stays yours.** The place, every object, the camera's every move and framing, the light, the colour and the focus stay as your clip has them — unless your words change them.
- **Nothing drifts in scale unasked.** What you did not ask to grow or shrink keeps its size.

## Changing the story

The same mind lets you go further than a new look. In ROUTIX's chat, you can change the story while you revise.

For example: in your clip, a man drinks a soda while he says his line. You tell ROUTIX: no soda — an eagle lands on his head and eats seeds. ROUTIX makes those moments anew around what happens now, and everything you did not name — the man, the place, the light, the camera, the pace — comes from your clip.

Or a dog on a beach stops, digs, and comes up with a starfish. A cyclist passing a café leans her bike on the wall and walks inside. A paper boat in a gutter lifts off the water and flies away. Any subject, in the same way: ROUTIX keeps no list of what a story may be about. (What may be made is set by our [Content Policy](https://routix.studio/help/policies/content-policy), for everyone.)

Why can it do this at all? Because of the mind behind it. A mind that is already imagining your whole scene anew — the place, the light, every subject in it — has only a small step left to imagine one of them doing something else.

In Studio, Optic keeps your story and changes what you name; a new story is the chat's. See [Change the story](https://routix.studio/help/chat/making-video-in-the-chat#change-the-story).

## How ROUTIX Blind thinks

Before it changes anything, ROUTIX Blind builds its picture of each shot the way that mind would: from the ground up.

How ROUTIX Blind holds a shot — built from the ground up, layer by layer:

1. **The base** — length, shape and speed — slow motion or real time
2. **Camera** — how it frames, moves and sees
3. **Physics** — shadows, reflections, gravity, cloth, dust and rain
4. **Light and atmosphere** — where light falls from, its warmth, the air, the hour
5. **Place** — the background, its depth, its colours
6. **Objects** — everything in between: where it sits, who touches it
7. **Subjects** — who and what is there — what never changes, what does
8. **Space** — what stands in front of what, and at what scale
9. **Time** — key moments, rhythm, what comes and what goes
10. **The whole scene, in words** — one description so full another mind could see it

- **The base** — how long the shot is, its shape, its speed: slow motion or real time.
- **Camera** — how it frames, how it moves, how it sees.
- **Physics** — where shadows fall and how soft they are, what reflects in what, which way things fall, how cloth and dust and rain move.
- **Light and atmosphere** — where the light comes from, how warm and strong it is, the air, the hour, the mood.
- **Place** — the background, everything in it, its depth and its colours.
- **Objects** — everything in between: where each one sits, and who touches it.
- **Subjects** — who and what is there, still or moving: what never changes about each one, and what does — an expression, a pose, a glance.
- **Space** — what stands in front of what, and at what scale.
- **Time** — the key moments, the rhythm, what comes in and what goes out.
- **The whole scene, in words** — one description so full that another mind could see the scene without the video.

Then, the change. It holds the whole picture still, goes to what you named, and changes only that — and what is new takes on the scene's own light, its camera, its colour, its depth and its scale. Everything else it leaves exactly where it was. It knows those things are there, and that they must stay.

## Two memories, and the new vision

ROUTIX Blind remembers in two ways. Its memory in words holds what it understood: every shot described, every fact about your project. Its memory in pictures holds what it has seen: frames, close-ups of each subject, the references you brought and the ones it made.

Now picture a film being told to a mind that once could see — and along the way, without a word of warning, a new vision is woven into the telling: the change you asked for. The listener hears no seam. The film that forms in that mind is one whole — the original and the new vision fused, with nothing out of place.

And because that mind builds the film piece by piece and keeps every piece, it can go back to any moment, build it again, and still know exactly what happened ten or twenty minutes before — where a sighted viewer, watching in the moment, loses whole sequences.

That is how ROUTIX Blind carries your characters, places and continuity through a film. Optic reads the whole stretch you chose at once, so a person or a thing keeps one identity in every shot; when its look changes, it is one subject with two looks — never two strangers. In the chat, after a build of many shots, a last check looks at every shot together.

## How ROUTIX watches a video

ROUTIX does not glance at your video. It watches it with its own video analyser — our own way of watching video, built for one purpose: to understand everything before anything is changed.

- **Every subject gets its own DNA.** Every person, animal, object or shape it finds gets its own identity: what makes it itself, the looks it takes on, and where it first appears.
- **What belongs to a subject has its DNA too.** A character's hair, a garment worn layer by layer, something held, a mark, a companion — each tied to its owner and followed moment by moment, wherever your request reaches it. In the chat, everything it finds gets its own record, tied to the subject it belongs to.
- **Everything that happens is kept as structure.** Every cut and every fade. What happens in each shot, and when. How the camera frames and moves. Where each subject is — whole, in part, or gone — checked frame by frame wherever there is doubt. Every word on screen, letter for letter.
- **Every sound, in the chat.** Every word spoken, with its moment and how it was said; the music, the room, the silences.
- **The whole video, fast.** It reads the whole stretch, and keeps what it read — so every step of every revision works from one understanding, never from a guess.

## Reasoning, again and again

ROUTIX Blind does not answer in one breath. It works the way a careful team works — its own agents, each owning its part:

- **ROUTIX**, the agent you talk to in the chat, plans and makes.
- **Fovea** reviews: every picture the chat makes is held up against your source and your request, and a take that does not hold up is sent back. See [Fovea](https://routix.studio/help/chat/fovea).
- **In Optic**, one reader takes in the whole stretch; another checks that reading against the frames themselves; another follows each thing you are changing, moment by moment — think first, then act. Everything is decided before anything is made, so your film is made once, from one understanding.
- **It asks for what is missing.** In the chat, ROUTIX asks — one thing at a time — rather than inventing what you did not say. In Optic, you can point with [Focus](https://routix.studio/help/focus/what-is-focus) at exactly the one you mean.
- **It tells you when it could not judge** rather than pretending it did.

In the chat you can give it longer to think: **How I think** offers **Blind**, **Blind · Considered** and **Blind · Deep** where the slider is shown — the longer stops take more time, and your credits stay the same. See [How I'll make this](https://routix.studio/help/chat/how-ill-make-this).

## One way of seeing, for everything

ROUTIX Blind keeps no list of what a video may show. A person, an animal, a product, a logo, a screen recording, a creature nobody has filmed before — each is read the same way, with the same care, by the same rules. There is no template for one kind of video and another for the next: one way of understanding, for every genre, style and subject. (What may be made is set by our [Content Policy](https://routix.studio/help/policies/content-policy), for everyone.)

## What it means for you

- **Characters stay themselves.** A person you replace stays the same person from the first shot to the last, through every look.
- **Changes land exactly where you point.** Every part of what you ask is tied to the thing it names — and with Focus you can show it the very spot.
- **Everything else stays yours.** What you did not mention comes from your video: the framing, the timing, the cuts, the light, the sound.
- **Change the story, not only the look.** In the chat, say what happens now, and the moments are made anew around it.
- **One press, and the rest is ROUTIX's.** In Optic, **Revise** is the payment; the reading, the plan and the film all follow inside that one run — and if the film cannot be made, every credit comes back.
- **Fewer surprises.** It asks for what is missing, and it says so when it could not judge.

## Where ROUTIX Blind works

- **In the chat** — behind every answer and every plan. The **Blind** side of the composer's switch is ROUTIX Blind seeing the work through from start to finish; see [Blind and Blink](https://routix.studio/help/chat/blind-and-blink).
- **In Optic** — behind the reading of your clip, the plan for your change, and the look at the finished film. See [What Optic does](https://routix.studio/help/optic/what-optic-does).

You never open ROUTIX Blind. You meet it in everything ROUTIX does.

## Frequently asked questions

### Can I use ROUTIX Blind on its own?

No. It is not a tool you open or a model you pick — it is the core the chat's agent and Optic think with. Every time you use them, you are using it.

### Why "Blind"?

Because it is designed to think the way a person who once could see, and lost their sight, rebuilds a world: from memory and description, layer by layer, without the noise of the pixels in front of it. Every character in ROUTIX's branding is blind for the same reason.

### Why "next generation"?

Because what stands behind Optic's revision reasoning is not one model and not one workflow, but a way of thinking: the mind of someone who once could see, rebuilding the whole scene — never patching over the old one.

### How do I judge it?

By the result. Revise a clip of your own and look at the quality; measure it against any revision model there is, or any that comes after. The judgement is yours.

### Can I change what happens in my video, not only how it looks?

Yes, in the chat: pick the moments, say what happens now, and ROUTIX makes them anew — everything you did not name comes from your clip. In Studio, Optic keeps the story.

### Does it work on any kind of video?

Yes — any subject, any style, any genre, read the same way. What may be made is set by our Content Policy, for everyone.

### Is the research published?

Not yet. Much of it stays with us for now; we plan to share more of it in time.

Source: https://routix.studio/help/why-routix/routix-blind · Updated 2026-10-24