Reviewing AI-assisted Unreal materials across quality tiers

Review an AI-assisted Unreal material by testing the result at the quality tiers your project actually uses. Define which part of the surface players need to recognize, compare the same view at each tier, and measure the change on representative hardware. A material that looks right in the editor can still lose an important cue in a packaged build.
Start with the part you need to preserve
Take an original model example: a stone switch on a wall. Its active state has a pale ring around the edge. At high quality, fine grain and small scratches make the switch fit the room. At low quality, those details can be candidates for simplification. The pale ring should still communicate that the switch is active from the distance where a player chooses whether to interact.
Write that distinction before asking an assistant to propose a material graph. Give it a reference view, the expected appearance at each tier and which inputs must behave the same. Ask for a small graph change that has a clear review surface. If it proposes removing a texture or a calculation, have it identify which layer of the material depends on it.
Check the actual graph branches
Epic's mobile performance guidance explains how a Quality Switch can choose different levels of material complexity. A simpler branch might use fewer texture lookups or fewer calculations. Inspect what each input branch does in your graph and make sure any default branch produces an intentional result.
Apply the same material instance to the same mesh for the comparison. Keep the view angle, lighting, exposure and distance stable. Then move the camera to the place where someone first needs to read the switch. A material preview gives you a close look at the surface. The gameplay camera tells you whether the material communicates what the player needs at the right moment.
The QORPO guide to comparing Unreal exposure covers a useful control for this step. If lighting changes between screenshots, the material change can appear to do work that belongs to the camera or exposure setting.
Track the cost that actually changed
Check the graph proposal against the performance question. If the goal is fewer texture reads in the lower tier, confirm that the relevant branch actually drops them. If the proposal adds several static options across material instances, review the permutations the project would compile. Epic's scalability reference describes that tradeoff for quality branches.
The editor can help you identify where to look. Pair its view with a capture from a packaged build on a representative device. Keep resolution, hardware and test route fixed. Record the settings you observed at runtime. If you need a procedure for checking settings in the running build, see how to find the effective Unreal console variable value.
A cheaper graph can still land in a scene where it barely affects overall frame time. Record its measured result at a scale that resembles the actual game. A room with a single switch and a corridor lined with the same material are different workloads.
What should the review record include?
Save the source material revision, the intended tier behavior, the packaged build, the actual device settings and one view from each tier. Add a brief observation about the cue. In the switch example, the key question is whether you can tell the active state at the approach angle. If a cold start or shader preparation affects the test, record the starting state so someone can recreate it.
When a branch loses the signal, narrow the edit. Keep the small visible marker and consider simplifying the layer around it. If an optimized version looks acceptable at close range but changes the readability of a distant interaction, the distant view belongs in the review record too.
If the result differs across two devices, keep their observations separate. Include the renderer and active quality level so the next reviewer can reproduce the difference before changing the graph again.
Where should a team start?
Pick one material already used in play, write down one cue that must survive and compare one branch. That gives the reviewer a small decision with a visible outcome. The AI-assisted texture guide covers a separate check at normal play distance, and the AI-assisted game code review covers evidence from a packaged build.
QORPO