How to make a Cloud in 3D | Houdini Tutorial

The Cloud Rig shelf tool can turn a modeled shape into a cloud volume. This lesson first applies it to a torus, then builds a more cloud-like silhouette from overlapping spheres, a flattened lower region, and smaller repeated forms. The geometry establishes the large shape before the cloud settings add volume detail.

This is a historical Houdini workflow using the Cookie and Copy nodes; the exact release is unverified. The recording demonstrates shape construction and a static cloud appearance, without simulating weather or building an animated cloud system.

Test the conversion with an unmistakable shape

At 00:01, create a Torus and adjust its radii so the center remains visibly open. In Cloud FX, choose Cloud Rig, select the object, and confirm. Houdini generates a cloud that follows the ring-like source.

The torus is deliberately not a typical natural cloud. It makes the relationship easy to inspect: the source geometry supplies the overall silhouette, while the rig supplies the volumetric appearance. The instructor notes that another modeled shape could be used as well.

Keep this first test separate from the later shaped example. It verifies the conversion process before adding complexity to the source geometry.

Build an uneven silhouette and cut the lower region

Around 01:02, create a Sphere and several Transform branches from it. Merge their outputs, then move and scale each branch to make an uneven group of overlapping lobes. The instructor changes heights, sizes, and offsets so the forms are not arranged in a perfectly straight line.

Using Transform branches retains a shared source shape while allowing separate placements. Check the combined outline through the Merge, but edit one transform at a time so it is clear which lobe is changing.

The desired design has a relatively flatter bottom and a more varied top. Add a Cookie node for a Boolean operation, connecting the sphere assembly and a Box that spans the lower region. Transform the box to cover the required width and depth, then change the operation away from Union to remove the unwanted lower portion.

The automatic captions garble the exact A/B subtraction wording, so the useful verification is the output geometry: retain the intended upper mass with its lower region trimmed. Do not blindly choose a subtraction direction from the caption phrase alone. The lesson is using this cut to art-direct the silhouette, not to establish a universal physical rule for all clouds.

Use scattered copies to break up the large forms

From 06:07, add Scatter to the shaped geometry. Lower the point count to a small working value—the demonstration begins around 11—then use a Copy node to place sphere geometry at those destinations. The sphere goes into the geometry-to-copy input; the scattered points provide the template positions.

The repeated spheres add smaller lobes to the broad form. Inspect the result before increasing count. If the overall shape is wrong, adding more copies only makes that wrong shape more complicated.

The instructor varies both point count and seed. Count changes how many repeated forms contribute to the result; seed changes their placement pattern. Try the seed while keeping the count fixed to compare different silhouettes without changing complexity at the same time.

The resulting assembly is relatively flatter underneath and more irregular above, which is the intended design for this example. Individual sphere radii and scatter settings are not presented as a universal cloud preset.

Convert the finished shape and refine volume resolution

Around 09:02, select the completed source geometry and apply Cloud Rig again. The rig creates the cloud from this more detailed silhouette. Name the source and generated objects clearly enough to distinguish them if additional clouds are added later.

Inside the generated cloud network, increase the Cloud node’s sampling divisions to obtain a more detailed volume. Wait for it to calculate and compare the result. This adjustment refines the volumetric representation; it is a different task from repositioning the original spheres to change the broad silhouette.

The rig also creates sunlight and skylight. Near 11:01, the instructor tests a warmer sunlight color and a cooler skylight contribution to change the cloud’s appearance. Lighting can improve the reading of the lobes, but it does not replace the source-shape work.

The final result is a static cloud ready for further scene development. No exact render-quality budget, animation, caching, or export procedure is demonstrated. For a related use of a Cloud Rig with much lower density, the depth-of-field and haze lesson uses a volume through the camera view to create atmospheric haze rather than a discrete cloud shape.