Unreal Engine 5.7 Preview

[controldirective (control)]

Spawning a larger number of Skinned Instanced Meshes (with the intended use as Nanite Foliage) seems to incur a huge performance penalty to the game thread, mostly “Post Tick Component Update” and “Transform or Render Data” but only during play/PIE, not in editor mode.

This makes actually using skinned meshes as foliage seemingly prohibitively expensive in game right now.

It happens with, or without a valid TransformProvider, with our without Dynamic Wind, it also happens when all animations are culled with r.Skinning.DefaultAnimationMinScreenSize 50.
Happens with a valid PCG spawner setup, happens also when spawned through BP.
It is not inherently due to instance count, the same meshes spawned as static instanced meshes perform fine.

Again though, this does not show up during regular in-editor use at all, but reliably during play/PIE.
Tested on earlier 5.7, 5.7 preview and a recent Github build, as of tonight.

Reproduced on a completely new empty project as well.

Repro steps:

Create new empty game project, enable PCG/PVE plugin, enable Nanite foliage, restart.
Drag in the default PCG graph (/ProceduralVegetationEditor/SampleAssets/PCG/Wind_Example_PCG.Wind_Example_PCG, using this one specifically for easier repro, it’s just a basic skinned instanced mesh spawner), set position to 0,0,0, set scale to 300/300/10.
This leads to approximately 35.000 points/instances.

(Snap to camera position 1, start PIE once, so the camera gets aligned to ground, to get same view in editor/PIE. :: )

That’s the comparison in this case, one image in editor, one in PIE with 35.000 instances, one in PIE with 85.000 instances. Adding more instances makes it significantly worse, I can make it reach up to 100ms in play, while it’s always essentially at 0 in editor. Worst performing image is with 85.000 points (graph scale set to 500/500/10).

Adding more instances further exacerbates the issue until you are most certainly game thread limited.

[jakubhaluszczak (Jakub)]

A few problems I’ve encountered:

VT textures in the Content Browser are wrongly displayed

It only affects sRGB textures like base color. Linear textures like normal maps, roughness etc.. are fine. Looks like an issue with gamma correction. This affects 5.7 and 5.6.

WPO on nanite voxels doesn’t work

When I use WPO on voxels, they don’t move but the material does. Basically unusable. I tested both my custom wind and the Megascans european beech trees.

Camera crop settings bug

1. Create a cine camera
2. Set crop settings to for example 1.33
3. Then click the arrow to reset it
4. Try to select 1.33 aspect ratio again

It doesn’t let you unless you manually change to no crop and then select 1.33 again. This affects 5.7 and 5.6.

HW lumen still doesn’t work out of the box on ISM’s with WPO

The same thing happened in 5.6. However, in raytracing if you disable WPO on nanite proxies (r.RayTracing.Geometry.NaniteProxies.WPO 0), or disable WPO on static meshes (r.RayTracing.Geometry.StaticMeshes.WPO 2), it works properly. It should be like that, or I have wrong settings or something?

Noticed that lumen is more accurate and slightly more stable with the same settings, so that’s great.
Also good to see GPU Visualizer back.

[Tina_Wisdom]

Previews are a great time to experiment with new tools - if you’re curious about it, give it a try! Just keep in mind that Previews are under heavy active development, so things may not be as stable or finalized as the full release iterations. Otherwise, there’s never any harm in trying something new and having fun with the engine!

[POTWD]

You might find this on your own later, but this video just came out this morning. The wind stopped in my test level as well. he describes what’s going on towards the end of his example. You can provide the “wind transform provider” asset to the skinned mesh spawner (template descriptor → transform provider) inside the PCG graph and get the wind back. Hope this works for you.

Unreal Engine 5.7 (Preview) – Nanite Foliage | Procedural Vegetation Editor Explained!

[COMBAS_REAPER]

Any plans to add more quality of life to level designers?

i personally dont use the 1234 keys to teleport or mark down areas, so my suggestion is

make it so i can change something’s pivot point with these keys or something else then i can change it’s keybinds myself. it’d be so much better for building stuff to switch between scaling and rotating.

another one, i am not sure if you’re able to achieve it due to how the engine is written, but i’d really love for when i scale it scales to one sides rather than from both, i either have to manually change it’s pivot point everytime when i wanna switch between dragging it, scaling it, rotating it. which isn’t very fun in my opinion.

make it so the normal scale to be one side, if i held down tab while scaling then it’d do both sides. please i need this, i know changing pivot points is a workaround but make it easier on us

[than_sad]

Hi, The suspension component of Chaos Modular Vehicles (Cluster Union Vehicle Component) isn’t working. Shows rigid, also changing the mass of body (Chasis) Until break the wheels. Changes in Max Raise, Max Drop and Spring Rate are visible in the viewport (increase or decrease space between body and wheels) but remain rigid.

Same results with Modular Vehicles example versions. And also trying to create new geometry collections from the body and wheels’ static meshes, I recreated the Cluster Union Vehicle Component from scratch with the same result: the suspension is rigid.

It looks like the changes in the chaos core affected the behavior of Chaos Modular Vehicles a lot.

If anyone else is also facing this problem, please share it to let me know I’m not the only one.

Best regards.

[Uncle_Fred]

There’s some great experimental features here, but Epic, please consider refocusing on engine performance, going forward. This is becoming a major reputational issue for the engine.

We really need to see UE achieve Source Engine levels of performance and clean visuals. This is far more important than feature development right now.

[Ibtesam_Sadiq]

1st Issue:
I am using ray-traced lumen with Megalights, and the lighting mode is set to Hit Lighting. When I set the r.RayTracing,Nanite.Mode =1.
It removes some meshes from the ray-traced reflection’s view, causing the dark areas and light bleeding. It was working fine on UE 5.6.

2nd Issue:
I am using lumen with lot of rect lights with temperature around 5500K. and megalights is enabled. scene in lumen looks little bit warm and dark compared to path tracer.
Either the path tracer is ignoring light temperature or Megalights is making the scene warmer. i am not sure why both look different. This issue is the same in 5.6.

[Ibtesam_Sadiq]

are you considering the implementation of Mega Geometry from NVIDIA? I think it will improve the ray tracing without costing a lot of performance and it will allow ray tracing on a large scale. We also won’t need Cvars like r.RayTracing,Nanite.Mode =1.

[Joshuab_artist (Joshua)]

yeah, true. I guess so. I’ll give it a try

[Jian.Ru]

Thanks for reporting the issue. As mentioned in the other thread, opaque hits from refracted rays are lit by hit-lighting. Unlike raster which uses screen probes, hit-lighting can only get indirect lighting from Lumen surface cache. Looking at your repro project, there are several issues. Firstly, the lids, labels, and pills are missing surface cache. If you do “r.Lumen.Visualize 1”, you can see they are yellow in the surface cache coverage tile. Those objects are small (in Unreal units) and get culled. You can lower r.LumenScene.SurfaceCache.MeshCardsMinSize which will prevent them from being culled. By default, only a single card is generated for the lid and label mesh. You can visualize it using “r.Lumen.Visualize.CardPlacement 1”. You can lower r.MeshCardRepresentation.MinDensity, rebuild the mesh, then it should bring some of the cards back. The top and bottom cards are culled by a different metric. Unfortunately, the threshold for that metric isn’t currently adjustable so there isn’t much you can do unless you are willing to change code. After reducing MeshCardsMinSize and MinDensity, those meshes should now have surface cache coverage but they will be mostly black. That is caused by radiosity probe density being too low. You can increase radiosity probe density by reducing r.LumenScene.Radiosity.ProbeSpacing at the cost of perf and indirect lighting update lag.

[Ibtesam_Sadiq]

I am having few issues with OpenScene:
BUG: landscape spline actor was broken and had gaps in the lumen surface cache scene view and HWRT reflection view in 5.6. but in 5.7 HWRT looks good but in surface cache landscape spline is still broken. issue resolves when i disable nanite on the source mesh.

500×500 2.84 KB

I am having few issues with open scene: BUG: landscape spline actor breaks and have gaps in the lumen scene view while looks fine in lit mode geometry wise but lighting will be incorrect. I have Nanite enabled on the mesh that i am using in landscape spline actor. But issue resolves when i disable nanite on source mesh. [Screenshot 2025-09-15 185931]

And I also can’t see foliage other than grass type in Lumen view modes.
I am using the following CVARS:
r.RayTracing.Nanite.Mode=1;
r.RayTracing.Culling=0;
r.RayTracing.Geometry.InstancedStaticMesh.Culling=0;
r.RayTracing.Geometry.LandscapeGrass=1;

[Ibtesam_Sadiq]

Thanks, I will try these cvars.
will it improve the GI inside glass with ray traced refraction?

[northstar]

5.7 might be the ‘stable’ engine i need… getting ready for nanite changeover to be finished with assets… like way back when 4.18 was the stable engine… i’m just about animations, but i need a stable landscape and rendering… but cameras can do wonders… finish nanite and char nanite and gpu and i’m good (

[Alex-G91]

I have the same issue with the official 5.7 Preview build on KDE Wayland.

[Krabworks (CG_Krab)]

There are still issues with Migration from 5.5 to 5.6 that break some blueprints. Any chance we will see these fixed?

Would like to move my project forward to 5.7 and beyond, but this is not possible if 5.5/5.6 blueprint projects are not fully forwards compatible yet.

This has been a problem since UE 5.4, fixed in a later update for 5.4 > 5.5.1 but Still NOT FIXED for 5.5 > 5.6 (looks like the problem was found for 5.4, fixed, then re-emerged and was not addressed in 5.5). More info and reproduction steps here: Some Inherited Blueprint Variables Breaks After Upgrade to UE 5.5 - #37 by hectavex

[kalmiya]

GameInput D-Pad is broken, at least for the PS5 DualSense Gamepad.

Repro-steps:

New UE5.7 Empty Project
Plugins: Disable “XInput Device”, Enable “Game Input Base” and “Game Input Windows” (restart)
Add “Input Mapping Context” IMC_Default
Add “Input Action” IA_Test

DefaultInput.ini add settings for the PS5 DualSense:

[GameInputPlatformSettings_Windows GameInputPlatformSettings]
bProcessController=True

[/Script/GameInputBase.GameInputDeveloperSettings]
+DeviceConfigurations=(DeviceIdentifier=(VendorId=1356,ProductId=3302),Description=“PS5 DualSense controller”,bOverrideHardwareDeviceIdString=True,OverriddenHardwareDeviceId=“DualSense”,bProcessControllerButtons=True,bProcessControllerSwitchState=True,bProcessControllerAxis=True,ControllerButtonMappingData=((1, “Gamepad_FaceButton_Left”),(2, “Gamepad_FaceButton_Bottom”),(4, “Gamepad_FaceButton_Right”),(8, “Gamepad_FaceButton_Top”),(16, “Gamepad_LeftShoulder”),(32, “Gamepad_RightShoulder”),(64, “Gamepad_LeftTrigger”),(128, “Gamepad_RightTrigger”),(256, “Gamepad_Special_Left”),(512, “Gamepad_Special_Right”),(1024, “Gamepad_LeftThumbstick”),(2048, “Gamepad_RightThumbstick”),(8192, “Gamepad_Special_Left”)),ControllerAxisMappingData=((0, (KeyName=“Gamepad_LeftX”,DeadZone=0.239532,bIsPackedPositveAndNegative=True)),(1, (KeyName=“Gamepad_LeftY”,Scalar=-1.000000,bIsPackedPositveAndNegative=True)),(2, (KeyName=“Gamepad_RightX”,bIsPackedPositveAndNegative=True)),(3, (KeyName=“Gamepad_LeftTriggerAxis”)),(4, (KeyName=“Gamepad_RightTriggerAxis”,DeadZone=0.239532)),(5, (KeyName=“Gamepad_RightY”,Scalar=-1.000000,bIsPackedPositveAndNegative=True))))
bDoNotProcessDuplicateCapabilitiesForSingleUser=True

Open IMC_Default, Add Mapping for IA_Test and map it to “Gamepad D-pad up”

Add Blueprint “PlayerController” BP_PlayerController
Add Blueprint “Character” BP_Player
Add Blueprint “GameMode” and

• set class defaults
  • Player Controller to “BP_PlayerController”
  • Default pawn to “BP_Player”

BP_PlayerController:

• add “Enhanced Input Local Player Subsystem”, drag out “Add Mapping Context” with mapping context set to IA_Default
• Connect Event BeginPlay to Add Mapping Context
• Add IA_Test to the blueprint
  • connect started to “print string” with in string “Started”
  • connect completed to “print string” with in string “Completed”

Run the game

• press digipad-up (keep it pressed), notice print of “Started”
• release digipad-up, notice print of “Completed”
• press digipad (keep it pressed), notice spam of "Started, Completed, Started, Completed, … "
  • expectation: It should print “Started” once, like the first time we tried it
• release digipad, spam stops
• Alt-tab to a different window. Try the above steps again, notice it works correctly the 1st time, and from that point on is in a broken state. You basically reset it on alt-tab so it works 1 time.

[Ibtesam_Sadiq]

I tried the suggested CVARS. There are improvements to the surface cache for small objects. thanks for sharing cvars.
But seems like there was a misunderstanding. I was pointing out objects (Pills) inside the glass. It gets darker only when I enable ray-traced refraction, but it looks fine with raster. And it also looks good with the path tracer.

[user_b7524e9b77c9fbd47ad1797454da6c7df2b8695d6c7063b06bff7f (丘木晨)]

Just replace the MI_Face_LacrimalFluid material

[The_Distiller]

Replace it with what exactly?