> 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/control-ai-output.md).

# Controlling AI output

How Intangible gives you real control over AI output – scene fidelity in Build, framing in Compose, model choice in Visualize.

Prompt-only generation gives you powerful output and no control surface: a character that drifts between shots, a composition you can't pin down, a client's product the model keeps re-interpreting. The render that comes back is someone else's idea of the brief.

Intangible's agents build the project on the canvas. When a result needs to be exact, you step into a mode and pull its lever by hand. Each mode is one lever.

* **Build locks the scene.** Objects, props, time of day, imported CAD, brand-true reference images. Everything that belongs in the shot, you place. The model honors what you brought.
* **Compose frames the moment.** Camera position, lens, aspect, aim target. Spatial decisions the model can't override.
* **Visualize picks the model.** Photoreal, stylized, video with first-and-last-frame interpolation. The auto-prompt is editable; the scene grounds it.

A render you can put your name on isn't a single prompt. It's three locked decisions.

## Which lever for which problem

The most common control failures with prompt-only tools, and where Intangible solves each:

| What you're trying to control                       | The right lever                               | Where to start                                                                 |
| --------------------------------------------------- | --------------------------------------------- | ------------------------------------------------------------------------------ |
| Same character or product across every shot         | Build (Image Reference)                       | [Image reference](/learn-more/concepts/image-reference.md)                     |
| Specific framing, lens, aspect ratio                | Compose                                       | [Camera controls](/compose/camera-controls.md), [Lenses](/compose/lenses.md)   |
| Photoreal vs stylized look                          | Visualize (model picker)                      | [Models](/visualize/ai-models.md)                                              |
| Lighting and time of day                            | Build (Sky) plus Visualize (Lighting Presets) | [Lighting presets](/visualize/lighting-presets.md)                             |
| Brand-specific style across the project             | Visualize (Custom Styles)                     | [Custom styles](/visualize/custom-styles.md)                                   |
| Specific 3D geometry the asset library doesn't have | Build (Smart Import or Generate 3D)           | [Generate 3D Asset](/build/generate-3d-asset.md)                               |
| A character or product the model has never seen     | Build (LoRAs)                                 | [LoRAs](/build/loras.md)                                                       |
| Reference image being ignored / livery not showing  | Build (use a generic object name)             | [Image reference](/learn-more/concepts/image-reference.md#what-overrides-what) |

## Why this works when prompting alone doesn't

Pure prompt tools render from text. Text is lossy. "A red sports car on a wet street at dusk" generates a plausible red sports car, not the 2024 Bronco the client briefed. The model fills in what it doesn't know, and the gaps drift between renders.

The one exception to "the fix isn't in the prompt": a too-specific object name *is* a prompt problem. The object's name flows into the final prompt, and a name like "Gulfstream G550" makes the model imagine its own version and override the reference image you attached. Keep names generic ("private jet") so the reference wins. The full precedence – geometry, then image reference, then the prompt – is in [Image reference](/learn-more/concepts/image-reference.md#what-overrides-what).

Intangible passes the AI model a structured 3D scene plus a prompt. The geometry, the placement, the bound reference images, and the camera framing are all there as structured input. The prompt is one piece, not the whole brief. See [How the visualizer thinks](/learn-more/concepts/how-the-visualizer-thinks.md) for the technical version.

The practical version: every Intangible page below is one of the levers. When a render comes back wrong, the fix is almost never in the prompt.

## Where to go next

<table data-view="cards"><thead><tr><th data-card-target data-type="content-ref"></th><th></th></tr></thead><tbody><tr><td><a href="/learn-more/concepts/image-reference.md">References</a></td><td>The single most-used technique for keeping a character or product consistent across shots.</td></tr><tr><td><a href="/learn-more/concepts/how-intangible-works.md">How Intangible works</a></td><td>The agents, the canvas, and the modes you step into to take control.</td></tr><tr><td><a href="/learn-more/concepts/how-the-visualizer-thinks.md">How the visualizer thinks</a></td><td>Why structured 3D plus prompt beats prompt alone.</td></tr></tbody></table>


---

# 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/control-ai-output.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.