> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runchat.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Generate new views with Atlas

> Turn one generated concept image into a 3D point cloud, then render it from new camera positions

A generated concept exists only from one angle. World Labs' Atlas estimates its 3D structure from that single image, then generates views from cameras you choose.

![A generated cardboard concept model and three new views of it generated with Atlas](https://mintcdn.com/runchat/ExzxwftWKPCOoCF3/images/examples/new-views-with-atlas.webp?fit=max\&auto=format\&n=ExzxwftWKPCOoCF3\&q=85\&s=895b3649357db47f219eaaf46c87372c)

> [card] [Open the example workflow](https://runchat.com/editor?id=93de02fa-302f-4ba6-b5c2-9d0641b53c9d)
>
> Make a copy of this workflow in your account to try it with your own images.

## How it works

1. **Generated concept (Flux 2 Klein)** makes the concept: *a cardboard architectural model by frank gehry*.
2. **Atlas point cloud (posed RGBD)** turns the image into a posed depth point cloud.
3. **Target cameras** is a list of camera positions (orbiting -35°, +35° and +70° around the model).
4. **Atlas: generate each new view** renders the point cloud from each camera, guided by a short description of the model.

## Tips

* Small orbits (under about 45°) stay most faithful. Larger moves invent more of the hidden side.
* For views from the back, try [3D models](https://docs.runchat.com/examples/new-views-with-3d-models.md) or [video models](https://docs.runchat.com/examples/new-views-from-video.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.