> For the complete documentation index, see [llms.txt](https://docs.m-xr.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.m-xr.com/marso-studio/preparing-your-assets/sourcing-and-preparing-assets.md).

# Sourcing & preparing assets

Marso Studio **enhances** materials — it doesn't repair geometry or compensate for a poor scan. The quality of your output is tied directly to the quality of your input. This page covers what makes a good input and where to find assets to try.

### If you're generating a scene

A scene starts from an image, not a mesh, so the rules below about textures and UVs don't apply — but the same principle does. **What you feed the scene determines what you get out of it.**

A good scene image:

* **Reads as a place, not an object.** Prompts that describe a set — what's on the bench, what's in the corner — decompose into more usable objects than prompts describing a single hero asset.
* **Has clearly separated objects.** Detection works on visual separation. A scene with distinct, individually visible props detects far better than one with a few large overlapping forms.
* **Gives each object room.** Every object is reconstructed from its own reference image, and 3D generation needs the object to fill most of that frame. Objects that are small, distant, cropped at the edge or mostly hidden behind something else tend to fail at the 3D stage.
* **Is clear rather than atmospheric.** Heavy fog, extreme lighting and motion blur all cost you objects.

If a detection comes back wrong, it's usually cheaper to fix the image than to keep re-running detection. See [Running detection again](https://docs.m-xr.com/marso-studio/reference/scene-generation.md#running-detection-again).

### Two questions that decide how to prepare your file

**1. Does it have a texture?**

Your asset must carry an image-based base-colour texture.

* ✅ Has a texture
* ❌ Vertex colours only — these aren't textures and will be rejected

**2. Where did the mesh come from?**

| AI-generated mesh                                           | Your own mesh                                  |
| ----------------------------------------------------------- | ---------------------------------------------- |
| Created with a generative model — text-to-3D or image-to-3D | An artist-made model, or a photogrammetry scan |

The preparation rules differ slightly for each.

#### AI-generated assets

* ✅ Has a texture
* ✅ **Has shading** — the model works best with assets that carry shading and lighting cues in the base colour; this outperforms flat shading
* ❌ Flat shading, or a plain base colour with no lighting information
* ❌ Already has PBR materials

#### Your own mesh

* ✅ **Unique UVs** — parts of the asset don't share the same UV space
* ❌ **Textures that tile across the model** — bake these down to a single UV map first
* ❌ **Procedural textures** — bake these to a single UV map first
* ❌ Already has PBR materials

A well-captured photogrammetry scan is ideal — clean geometry, well-defined surfaces, no mesh "boiling" or sloppy reconstruction artefacts.

{% tabs %}
{% tab title="Good Mesh" %}

<figure><img src="https://515487347-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1qUuQhJ8sb2nM5XkrXfN%2Fuploads%2Frnrg12GcyH47JUG7Khbj%2Fimage%20(21).png?alt=media&amp;token=900842e6-5c7b-4c7e-9046-b4cb43353820" alt="" width="375"><figcaption></figcaption></figure>
{% endtab %}

{% tab title="Bad Mesh" %}

<figure><img src="https://515487347-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1qUuQhJ8sb2nM5XkrXfN%2Fuploads%2FqhXApLXrqSnquGFsgwEW%2Fimage%20(22).png?alt=media&amp;token=83eaec6f-f079-40bf-ba7f-f3d51a393946" alt="" width="375"><figcaption></figcaption></figure>
{% endtab %}
{% endtabs %}

> **Why "no existing PBR"?** The model's job is to *predict* the PBR stack. An asset that already carries PBR materials works against that prediction — bring a clean, base-colour asset and let Marso Studio do the rest.

### Keep files under 100 MB

Large files are slow to upload and more likely to fail. Decimate or optimise heavy meshes before uploading.

### Check before you generate

* Confirm the asset opens with its base-colour texture visible.
* Inspect the UVs — look for overlapping or shared UVs on parts that should be unique.
* Make sure there's no PBR already applied.

### Where to find assets to try

* **Example assets** — click an empty **Load 3D Model** node and pick one from **Example Assets**, so you can see how Marso Studio works before bringing your own. A few also show on the node itself, under *Or start with a demo asset*.
* **Sketchfab** — plenty of downloadable `.glb` assets. Two collections of ours work well with Marso Studio: [Auto PBR Results](https://sketchfab.com/m-xr/collections/auto-pbr-imagetomaterial-4137cee2eca94d4b9eb3376dbac359a0) and [Test Data](https://sketchfab.com/m-xr/collections/img2mat-test-data-ed113062403d410f85138cffdac018fc).
* **Generate a scene** — the fastest way to get a lot of assets at once. See [Generate Entire 3D Scenes](https://docs.m-xr.com/marso-studio/quickstarts/generate-entire-3d-scenes.md).
* **Generate a single asset** — one asset from a prompt or an image. See [Generate a PBR 3D Model](https://docs.m-xr.com/marso-studio/quickstarts/generate-a-pbr-3d-model.md).

### Evaluating a mesh on Sketchfab

Switch to MatCap view mode before you download. It strips the textures so you can inspect the geometry directly: you want clean, well-defined surfaces without noise, holes or overlapping faces.

<div><figure><img src="https://515487347-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1qUuQhJ8sb2nM5XkrXfN%2Fuploads%2FbOspaooexFcX1MA5gkmz%2FModelInspect.jpg?alt=media&amp;token=a8b3fc5a-8dc5-4072-87a8-f0aff0ec1f7f" alt=""><figcaption></figcaption></figure> <figure><img src="https://515487347-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1qUuQhJ8sb2nM5XkrXfN%2Fuploads%2F5T6DlbVbTU4Osbt9nFDR%2FMesh.jpg?alt=media&amp;token=bbdec547-9140-465e-88ad-b16c4b33523f" alt=""><figcaption></figcaption></figure> <figure><img src="https://515487347-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1qUuQhJ8sb2nM5XkrXfN%2Fuploads%2FIMXV6BY41jeUXb2rciUX%2FMatCap.jpg?alt=media&amp;token=5442358f-b6bb-442f-9164-df86d11fa809" alt=""><figcaption></figcaption></figure></div>

***

Next: check the exact format rules in [3D file support](https://docs.m-xr.com/marso-studio/preparing-your-assets/3d-file-support.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://docs.m-xr.com/marso-studio/preparing-your-assets/sourcing-and-preparing-assets.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.
