Building Mesh-Free Portal Bubbles in UE5 with Post-Process

Many spatial effects in real-time graphics usually rely on geometry: separate shells, meshes, or additional volumetric forms. I became curious whether I could take a different approach and create a similar effect without any auxiliary objects at all. I wanted to make spheres that would read as semi-transparent volumes with depth, rather than just a screen-space mask, while not requiring a mesh.

This project grew out of my previous research on synchronizing SceneCapture and the camera in Unreal Engine 5. In the previous article, I focused on the core principle of aligning the camera and capturing images, but here that principle becomes part of the effect itself. Everything depends on the image used inside the sphere. If it is the current scene, the sphere reads as a transparent volume embedded in the level. But if the image comes from a synchronized SceneCapture, it becomes a system of stylized "bubble portals" that reveal another part of the level while remaining transparent to one another where they intersect.

At the core of the effect is a function that determines the current pixel's position relative to the spherical volume and uses that position to reconstruct the normal for the visible part of the sphere. This is what makes the bubble read as a volumetric object rather than just a mask layered over the frame. Everything else in the effect is built on top of that foundation.

With a single sphere, the setup was relatively straightforward. But as soon as multiple spheres appeared in the frame, new problems emerged. In the overlap areas, the sense of volume started to break down: instead of transparent bubbles, the result looked like a standard layered overlay, where one sphere simply covered another. What I wanted, though, was for them to feel like transparent yet volumetric objects existing in the same shared space.

That meant I had to tackle sphere sorting separately, along with careful handling of overlap and transparency between them. This is what made it possible to preserve the readability of the effect even in more complex intersection areas.

It was important to me that the entire effect existed specifically as a post-process system. This becomes especially clear when switching between different view modes: you can immediately see that the bubbles are not ordinary scene objects, but are generated entirely through post-process logic.

To compare the performance of the mesh-based and post-process approaches, I built a separate simplified test scene and evaluated them under the same conditions. In my test, the post-process version turned out to be slightly lighter in terms of GPU time, required fewer draw calls, and used noticeably fewer primitives. At the same time, it completely avoided the RenderTranslucency cost that comes with mesh-based spheres, although, as expected, part of that cost shifted into RenderPostProcessing.

The difference is also very clear in the debug views. In Shader Complexity and Overdraw, the mesh-based version behaves like standard transparent geometry, while the post-process spheres do not produce the same kind of translucent layering in the areas where they intersect.

At the same time, I wanted the system to work as a flexible artistic tool. That is why these "portal bubbles" allow control over edge softness, transparency, the character of the image inside, and the balance between color, density, and distortion. Because of that, the same effect can feel almost ghostlike and barely noticeable, or, on the contrary, dense, vivid, and visually rich.

These spheres can also be integrated quite naturally into the logic of a game world. They can function as magical zones, spatial anomalies, or markers of special world states. They work especially well in scenes involving dreams, memories, spatial distortion, or transitions between different environmental states. In the end, this project became, for me, a way of exploring how far post-process can be pushed toward genuinely spatial effects.

Alina Ledeneva, UE5 Technical Artist

Read the original publication here