> 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/chat-agents-intangible/in-the-modes.md).

# What the agents do in Build, Compose and Visualize

What to ask Intangible agents for from the canvas or Build, by mode: environments, populators, roads, objects and poses; camera moves and reels; keyframes, videos and looks.

Ask from the canvas or Build for a camera move, an animation or a render, and the work lands on the shot. On the canvas the agents run whole jobs; for each mode they are the fastest way through the slow parts of its work. The scenes, models and cameras they make are ordinary scene content, so you can open Build or Compose and refine them by hand.

## Watch

{% embed url="<https://www.youtube.com/watch?v=XxbxqcGCuR4>" %}

![The chat panel in Build: the request "Add a wooden chair on the sidewalk to the left of the car." and the reply saying the agents added a wooden dining chair called Wooden Chair, checked it from two views and kept it clear of the lamppost. The chair stands selected in the viewport](https://1179478100-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FSWrjIqlgVAblx6vHGJiM%2Fuploads%2Fgit-blob-20df7f59d02fddcdf48aae5849fcf31d1c262889%2Fchat-build-01.png?alt=media)

## Build

If you have staged a previs scene before, you know the shape of the work: an hour collecting forty pieces of vegetation for a forest, then another hour arranging them so it looks like a forest and not a plant nursery. Prompt for the environment instead, "create a tropical forest", and the agents collect the assets and arrange them, ready for you to nudge.

1. **Prompt for an environment.** "Create a city block." "Build a wedding scene." "Make a tropical forest, roughly 100 by 100 meters." Specific prompts get specific results.
2. **Wait for the result.** Environments come back as a populator (a container that distributes assets across an area) or as individually placed assets.
3. **Adjust the populator.** Ask for changes in the chat panel ("make the grass sparser", "more palm trees, fewer banana trees"), or click it in the viewport to open its contents and change it by hand.
4. **Add objects.** "Add a character on the left side of the cabinet." "Drop a coffee table between the two couches." The agents read spatial words like left, right, between and behind, and try to honour them.
5. **Delete an object.** "Delete the rightmost couch." Name it the way the scene names it.
6. **Pose a character.** "Make the character stand." "Rotate the character to face the camera." Name your characters if you have several.

### Paths, roads, and anything along a curve

The phrasing decides what you get:

* **Things along a curve.** "A winding stone path with lampposts every 10 meters." "Line this driveway with pine trees on both sides." The agents create a [spline populator](/build/spline-populator.md): a curve with assets distributed along it. Several things on the same curve come back as one populator with several elements.
* **A solid surface.** "Add a curved road as one solid surface." The agents extrude one continuous mesh along the curve. Use this for roads, paths and ribbons that need to read as one surface in renders.

### Working with character names

For a character that must stay consistent across shots, give it a plain name and put an [image reference](/learn-more/concepts/image-reference.md) on the character object instead of relying on a description.

### Custom assets

The agents find your imported models by their names and descriptions. A model named "2024 Bronco, matte olive" answers to "drop in the green Bronco"; one named "object-12" will not. Name and describe models well when you import them. See [Smart Import](/build/import-your-own-models.md).

## Compose

The chat panel is on the canvas and in Build; Compose and Visualize have none, so ask from the canvas or Build. Describe the motion and the agents author keyframes for it:

* **Object motion.** "Move the cube 3 meters left over 2 seconds."
* **Camera moves.** "Orbit around the product for 180 degrees." Use director vocabulary: orbit, dolly, push in. Each camera-move prompt creates a new shot with its own camera.
* **Whole reels.** "Create a cinematic reel of the scene" produces a series of animated shots.

![A generated cinematic reel: a list of named shots from Opening Crane Up to Grand Finale Dolly Out, each an editable camera move](https://1179478100-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FSWrjIqlgVAblx6vHGJiM%2Fuploads%2Fgit-blob-a22a6d9ef40b8f23334ca144890f80742211d213%2Faic-animation-01.png?alt=media)

The keyframes land on the scene timeline as ordinary keys. Refine them in Compose like anything you keyframed by hand. See [Animation](/compose/animation.md).

Three behaviours to know:

* **Each camera-move prompt makes a new shot and camera.** To change a generated move, clear its animation and re-prompt rather than stacking prompts on the same shot.
* **A move on one subject follows it.** When a camera move frames a single object, its aim target is parented to that object, so the camera keeps tracking it if the object moves. A move framed on several objects aims at a fixed point. To follow one of them, use [Aim Camera and Target](/compose/aim-camera-and-target.md).
* **A full turn of an object needs keyframes in between.** Two keyframes with the same orientation have nothing to interpolate, so ask for the rotation in segments. Camera orbits already get a keyframe about every 45 degrees.

## Visualize

Ask from the canvas or Build for a render, a video of a shot, or a new look:

* **Images.** "Render keyframes for shots 01 to 05."
* **Videos.** "Render shot 02 as video from its keyframe."
* **Looks.** "Use the golden hour lighting preset on every shot in scene 01."
* **Shot prompts.** "Write a prompt for shot 02." The agents describe what the shot's camera sees, from the camera and the scene: the setting, the framing, and each subject and where it sits. Ask them to render it and they use that description in place of the [auto-prompt](/visualize/auto-prompt.md).

The agents check which models, styles and presets a shot can use, then render or apply them. They pick the image model themselves (GPT Image 2.5 Sunburst for new shots) unless you name one in your message: "Render shot 02 with GPT Image 2." They write the settings into the shot's Visualize panel, so you can open the shot, see exactly what was used, change it and render again yourself.

{% hint style="warning" %}
In the Visualize panel, a large video shows a notice with its credit count above **Generate**. A video the agents start goes straight to the job, so you never see that notice. Before they start, the agents say what you'll get and its credits, and wait for you to agree. Check that figure before you do.
{% endhint %}

## Images stay in the conversation

The images you upload and the screenshots the agents take while they work are carried into later turns with the text. Earlier turns stay as written rather than being compressed into a summary, so they work from what you actually said. In a long conversation the oldest turns drop out of what the agents read, so repeat a reference that matters late in a project.

## What they are good at

| Capability           | What to ask for                                        |
| -------------------- | ------------------------------------------------------ |
| Environment creation | Cities, forests, interiors, beaches, urban districts   |
| Asset addition       | One object, placed with spatial language               |
| Asset deletion       | "the rightmost couch", "the dragon", "everything red"  |
| Character pose       | Standing, sitting, running, gesturing, custom poses    |
| Population scaling   | "More vegetation", "fewer cars", "double the crowd"    |
| Spline scattering    | Paths, fences, tree lines, lampposts                   |
| Road generation      | A curved road or path as one solid surface             |
| Object animation     | "Move the cube 3 meters left over 2 seconds"           |
| Camera moves         | "Orbit around the product for 180 degrees"             |
| Renders and looks    | Keyframes, videos, style and lighting presets per shot |

For habits that get better results, see [Tips for directing agents](/chat-agents-intangible/limits.md).

## Related

* [Chat](/chat-agents-intangible/chat.md)
* [Asset library](/build/asset-library.md)
* [Populators](/build/populators.md)
* [Image reference](/learn-more/concepts/image-reference.md)
* [How the visualizer thinks](/learn-more/concepts/how-the-visualizer-thinks.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/chat-agents-intangible/in-the-modes.md?ask=<question>&goal=<user_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.