The main rule
A Merge node combines data that already belongs to the same compatible system. A Render node usually converts a richer system—Shape, particles, 3D or USD—into a pixel image that can re-enter the regular 2D pipeline.
That simple distinction solves a large share of confusing Fusion connections. Similar names do not mean interchangeable data types.
Normal Merge and MultiMerge
The normal Merge works with image layers: Background, Foreground and an optional mask. MultiMerge extends the same image-compositing idea when many 2D layers need to be managed together.
These nodes belong near the final image assembly stage, after any specialized Shape, Particle or 3D branches have been rendered to images.
Merge3D, SMerge, pMerge and UMerge
Merge3D combines classic 3D scene elements. SMerge combines vector Shape elements. pMerge combines compatible particle streams. UMerge combines USD scene data.
Each one is useful inside its own world, but none of them automatically converts that world into a normal image. That is why a separate render boundary still matters.
Render nodes as system boundaries
SRender turns Shape data into pixels. pRender turns a particle simulation into pixels. Renderer3D renders a classic 3D scene. URenderer handles the USD-oriented scene output path.
After that conversion, ordinary image-processing tools can work again. This is the point where a specialized branch can be merged over footage, graded and delivered.
When to render early or late
Render late when you still need to change object-level behavior inside the specialized system: particle forces, 3D transforms, Shape geometry or USD hierarchy. Keeping the native data alive preserves control.
Render earlier when the branch is finished and the rest of the work is purely 2D. This can simplify the Flow and sometimes improve performance, especially when a heavy subsystem no longer needs to be recalculated.
How to read Merge and Render boundaries in a Flow
When a graph becomes complicated, identify every Merge-like node and ask what data it is combining. Regular Merge combines images, Merge3D combines scene objects, SMerge combines Shape data, and particle or USD merge tools operate inside their own systems.
Then locate the render boundaries. SRender, pRender, Renderer3D and the USD renderer are conversion points where specialized data becomes an ordinary image. Everything after that point can usually enter the familiar 2D compositing pipeline.
This reading method is faster than inspecting every parameter. First determine the data domain, then the merge operation inside that domain, then the render boundary that hands the result to the next domain.
Practical example: several systems meeting in one composite
Imagine a shot with a Shape-based lower third, a particle spark pass and a 3D logo. Each branch is built in its own native system: Shape → SRender, particles → pRender, and 3D scene → Renderer3D.
The three rendered images can then enter the regular 2D pipeline and be layered over MediaIn with separate Merge nodes. Color, blur, glow and grain matching can be applied after each render or during the final assembly as appropriate.
This structure keeps the graph modular. A change to particle behavior stays inside the particle branch, a camera adjustment stays inside 3D, and the final Merge chain remains responsible only for image compositing.