> For the complete documentation index, see [llms.txt](https://help.intangible.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.intangible.ai/learn-more/concepts/how-intangible-works.md).
# How Intangible works
You describe a project and bring your files. Intangible agents plan it, build one 3D scene on the canvas and render from it. You step into Build, Compose or Visualize when you want the controls.
You describe what you want to make and bring what you have. Intangible agents plan the job, build a 3D scene from it and render your shots from that scene. You decide what gets made, and you take the controls whenever you want precision.
## What you bring
A project starts on Home with a prompt and, if you have them, files: photos of the real product, location or people, PDFs such as a script or a spec sheet, and 3D models or scans. Everything you bring gets used. Everything you leave out, the agents generate. The more specific your inputs, the less they guess. See [Start a project](/get-started/start-a-project.md).
## The agents and the brief
Intangible agents are the AI built into Intangible. You direct them in the [chat panel](/chat-agents-intangible/chat.md) in the bottom-right corner. It sits on the canvas and in Build, with your own conversation on each project shared between the two. Close it and it folds to a bar at the bottom right that reads CHAT over the conversation's title; click the bar to reopen the same conversation. Camera moves, keyframes and renders are things you ask for there too; the agents carry them out in the scene.
With **Guide me** on, the agents ask a few questions, then write a brief: the subjects, the look, the scenes and shots. You confirm it once, and that confirmation starts the build. With **Skip**, they skip the questions and build straight from your prompt; a brief can still land on the canvas as they build. See [The brief](/canvas/the-brief.md).
From there the agents fill in what is missing: a reference sheet for each subject group, the set with its models inside it, motion where the brief calls for it, and a framed shot for every shot in the brief. With Guide me, they then stop and ask what to render; nothing is rendered until you choose. Messages to the agents use credits, as do the references and renders they make. See [Credits and tokens](/learn-more/concepts/credits-and-tokens.md).
## The canvas is home base
The canvas is where a project opens and where the agents put their work. The brief, your uploads, the subject groups with their references, each scene with its shots, the renders and the asset packs all sit on it as cards. You can trace any render back to the scene, the shot and the references it came from.
Press **0** for the canvas from anywhere. See [The canvas](/canvas/canvas.md).
## One 3D scene under every shot
Between what you bring and what you get there is one 3D scene. Every still and every video renders from it, through a camera placed in it. Three things follow from that:
* **Subjects stay consistent.** The pod in shot 01 is the same model, with the same reference, as the pod in shot 05. The image model is not reinventing it per frame, so it does not drift.
* **Scale is real.** The scene is laid out at true scale. A 1.2-meter rover sits next to a 5.4-meter pod in the right proportion, on a crater the right size.
* **Cameras are real.** Each shot is an actual camera with a lens and a position. A 50mm on the release and a long lens on the bounce frame the way those lenses do, and the framing stays put when you change the look.
Motion lives in the scene too. When the agents simulate a landing, the video rendered from that scene moves the way the simulation does, instead of the way a model guesses it should. See [How the visualizer thinks](/learn-more/concepts/how-the-visualizer-thinks.md) and [Keyframes to video](/canvas/keyframes-to-video.md).
## The scene is the source
The 3D scene is what you work on. Renders are what it produces. When you change something, you change what builds the image, not only the image.
* **Change the scene and future renders follow.** Move a prop, change an object's reference image or add a camera, then render again. Every render from then on uses the new scene. Renders you already made stay as they are, so you can compare.
* **Angles stay consistent.** An object in the scene carries its 3D geometry, a description and an image reference. A new camera on the same object renders it from the same model, the same description and the same reference, so it stays right. Editing a flat render instead tends to drift, because nothing tells the model what the other side looks like. See [Object details](/build/object-details.md).
* **It runs in the browser.** Full 3D control, cameras and lenses included, with nothing to install.
```mermaid
flowchart LR
subgraph S[One 3D scene]
O1[Object: geometry + description + reference image]
O2[Object: geometry + description + reference image]
end
S --> C1[Shot 01 camera]
S --> C2[Shot 02 camera]
S --> C3[Shot 03 camera]
C1 --> R1[Render]
C2 --> R2[Render]
C3 --> R3[Render]
E[You change the scene] --> S
S -. future renders only .-> R4[Next render]
```
When something looks wrong, fix it where it starts: the brief if the idea is off, the scene if the problem repeats across shots, the output if it is a one-off in one frame. See [Troubleshooting Intangible agents](/learn-more/faq/troubleshooting-the-agent.md).
## The modes are where you take control
Build, Compose and Visualize are the hands-on tools for the same project. Open one from a card on the canvas, or press **1**, **2** or **3**:
| Mode | You change | Open it from |
| ------------------------------------ | ------------------------------------------------------------------- | ------------------------------------------------ |
| [Build](/build/build.md) | The set: objects, references attached to them, environment, terrain | A scene card, **Edit in Build** |
| [Compose](/compose/compose.md) | The shots: camera, lens, aspect ratio, timing, animation | A shot card, **Edit in Compose** |
| [Visualize](/visualize/visualize.md) | The render: prompt, model, style, lighting | A render made from a shot, **Edit in Visualize** |
Use them in any order. The agents' work and yours land in the same scene, and what you do in a mode shows on the canvas. What they build in the scene is ordinary scene content, so you can change it by hand in Build or Compose.
One rule catches people: animation belongs to the scene, not to a single shot. Keyframes record on the shared scene timeline while a shot is active, and every shot whose time window covers that moment shows the motion. See [Animation and shot time](/learn-more/concepts/animation-and-shot-time.md).
```mermaid
flowchart LR
A[Prompt and files on Home] --> B{Guide me?}
B -- yes --> C[Brief: you confirm]
B -- Skip --> D
C --> D[Agents build the 3D scene on the canvas]
D --> E[Renders: stills and video]
E --> F[Asset pack]
D <-. you, by hand .-> M[Build / Compose / Visualize]
M -.-> E
```
## What you get out
* **Stills:** a keyframe for every shot with [Batch visualize](/canvas/batch-visualize.md), or one shot at a time in Visualize.
* **Video:** shots rendered with the scene as the motion reference. See [Keyframes to video](/canvas/keyframes-to-video.md).
* **3D models:** the scene exports as GLTF or USD for other tools. See [Export formats](/publish/export-formats.md).
* **Asset packs:** one ZIP of the cards you drag onto the pack, such as renders, videos, references and models. See [Asset packs](/canvas/asset-packs.md).
## Related
* [Your first project](/get-started/your-first-project.md) – the whole loop, end to end.
* [Projects, scenes and shots](/learn-more/concepts/projects-and-scenes.md)
* [References](/learn-more/concepts/image-reference.md)
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.
Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:
```
GET https://help.intangible.ai/learn-more/concepts/how-intangible-works.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
