Epic Games - MetaHuman Animator Markerless Motion Capture Plugin

[alizavi]

im fixed this problem , just update epic games launcher and retry to install plugin.

[AJMarrento (A.J)]

Is anyone else getting the error ‘GPU hung or D3D device removed’ when pressing the Process button? Afterwards, UE5 crashes and closes

I’ve been rummaging through forums, asking AI, for 2 days, and nothing solves the problem

My GPU: XFX SPEEDSTER QICK 319 AMD Radeon™ RX 6700 XT BLACK Gaming Graphics Card with 12GB GDDR6 HDMI 3xDP, AMD RDNA™ 2

[Ciprian_Stanciu]

I had this error and just realized how to fix it : close all unreal editors

[viktosw]

Looking at logs, does it mean its some sort of driver issue?

I’m having the same error, I checked hardware requirements and it points to 6800xt but I have seen people having the same error with 6900xt. I’m not sure if its about the hardware requirements or not

[Ciprian_Stanciu]

I think the requirements for this plugin may be a little beefy on the GPU side, specifically VRAM. I just did a small test with a 10s clip on my 7900 XTX and compared to before processing, when pressing process it consumes an extra 3GB of VRAM (from 9 to 12 in my case).

[Truth2]

Hi,

I’m testing the Markerless Motion Capture plugin on UE 5.8 and experiencing very slow processing speeds.

My setup:

• GPU: NVIDIA GeForce RTX 3070 Ti (8GB VRAM)
• RAM: 47.7GB
• OS: Windows 11
• UE: 5.8
• Plugin: installed to engine (not project)

Issue:
Stage 2/5 takes extremely long to complete. Processing 300 frames takes over 15 minutes.

During processing, GPU usage shows 50-100% (3D), so GPU is being used. However the speed seems far below what would be expected.

What I tried:

• Converting iPhone footage (1080x1920, 30fps fixed) to JPEG sequence via FFmpeg before ingesting

Both gave slight improvement but processing is still very slow.

Is this expected performance for RTX 3070 Ti, or is there a known issue with processing speed?

Any advice would be appreciated.

If anyone has found a way to improve processing speed (driver settings, project settings, or any workaround), please share. I would really appreciate any suggestions.

[Kinos141]

Anyone got this working on Linux?

Keep getting this error:

EDIT: figured it out.

Run “Path/th/engine/Engine/Build/BatchFiles/Linux/Build.sh UnrealEditor Linux Development”

This will build all plugins including the Markerless plugin.

Worked like a charm.

[Trinary_Games]

Update: Finally got it to install by using the Fab plugin in the editor for 5.8 (in the Toolbar under Window > Fab). So if anyone is still having trouble I would try to bypass Epic Games Launcher and use the editor version of Fab. If Fab doesn’t show up under Window > Fab, your install of 5.8 might be corrupted or out of date (mine was). I deleted the 5.8 install, reinstalled it, started a new ‘Getting Started’ project, opened up Fab from the editor, I did NOT log in to my account, search for the plugin and add it to the project.
Note that this method DOES NOT install the plugin globally (to all UE5.8 projects), so you will have to navigate to your new project from above in Windows Explorer: PathToYourProject > Plugins > Marketplaceand copy/paste MetaHumanBodyTracker folder to your 5.8 engine folder: PathToYourInstall > UE-5.8 > Engine > Plugins > MetaHuman

-- –

Wanted to add my voice here that the plugin does not download/install for me either (Windows 11). I have “An unknown error has occurred during installation. Error Code: IS-EO-Lnc-BldInfo:Http-403” like some other users. Additionally I was able to get “Could not download installation information. Please try again later. Error Code: MD-0011-0” one time.

I was hoping that meant there were fixes coming down the pipe, but I am now back to the first error. I already tried restarting everything and Smart App Control is disabled.

[tcsyoussef]

In my case**

Clear the Epic Games Launcher Cache**
Corrupted temporary files are the most common culprit for Http-403 errors. Clearing them forces the launcher to grab fresh data.

1. Press Windows Key + R to open the Run dialog box.

2. Type %localappdata% and hit Enter.

3. Open the EpicGamesLauncher folder.

4. Open the Saved folder.

5. Delete the webcache folder (you might also see webcache_4147 or similar—delete those too).

   Sign Out and Sign Back In

[RobertSfarzo]

(post deleted by author)

[ZenPigeonGames]

Yeah same issue, amd broke something for all 6xxx cards in the drivers past 24.6.1 and never fixed it.
If you revert to 24.6.1 the plugin will work just fine, but you’ll have to comment out a check in D3D12RayTracing because it will hit every frame complaining about the drivers being too old and not supporting something.

I have a 6750XT and it can process a 3 seconds animation in ~2 minutes perfectly fine.

Make a custom profile for the fans otherwise with this heat you’ll hit the hotspot max temp, I run it with with max freq 80%, Voltage 90% and aggressive fan curve, this way the hotspot doesn’t go beyond 65c.

[AJMarrento (A.J)]

More information:

My video has already been formatted correctly (Horizontal, 1920x1080, constant 30 FPS). I have already tried increasing Virtual Memory and performing a clean driver installation, but the issue persists.

My Specs:

GPU: AMD Radeon RX 6700 XT

OS: Windows 11

The log shows that NNERuntimeORT throws an input shape error, followed immediately by a D3D12 GPU crash (DXGI_ERROR_DRIVER_INTERNAL_ERROR) when the DirectML execution provider tries to run.

Here is the relevant part of my crash log:

[LogNNERuntimeORT: Error: Input shape does not match input tensor input of index 0]
[LogMetaHumanPipeline: Model variants resolved - Detector:2, Headpose:0, Solver:0]
…
[LogD3D12RHI: Error: GPU crash detected: - Device 0 Removed: DXGI_ERROR_DRIVER_INTERNAL_ERROR]
…
[LogNNERuntimeORT: Error: E:_work\1\s\onnxruntime\core\providers\dml\DmlExecutionProvider\src\DmlCommandRecorder.cpp(371)\onnxruntime.dll! … Exception(4) tid(18ec) 887A0020 Um problema interno impediu o driver de executar a opera??o especifica. O estado do driver provavelmente ? suspeito, e o aplicativo n?o deve continuar.]
…
[LogD3D12RHI: Error: hr failed with error DXGI_ERROR_DEVICE_REMOVED with Reason: DXGI_ERROR_DRIVER_INTERNAL_ERROR]
…
[LogMetaHumanPipeline: Error: Process error in node “RealtimeMonoSolver” on frame 0 - Code 1, Message “Failed to face detect”]

[AJMarrento (A.J)]

Hi everyone, I found a temporary solution.

If you are using an AMD Radeon RX 6000 series card (like the RX 6700 XT) and experiencing consistent crashes or the “D3D Device Removed / GPU Crashed” error at Frame 0 when running MetaHuman Performance, here is the definitive step-by-step guide to fixing it.

The Root Cause
Recent AMD Adrenalin drivers introduced a breaking conflict with how Unreal Engine 5’s Neural Network Engine (NNERuntimeORTDml) communicates with the hardware via DirectML. When MetaHuman tries to process facial tracking, the driver panics, freezes the Slate UI, and Windows forces a driver reset to protect the system.

STEP-BY-STEP FIX
Step 1: Before tweaking drivers, make sure the AI can actually read your source file:

Step 2: Downgrade to AMD Driver 24.6.1 (Crucial)
The community found that the DirectML breakdown started happening after mid-2024. Reverting to a highly stable previous version completely fixes the AI computation breakdown.

Download the Adrenalin 24.6.1 installer directly from the official AMD Previous Drivers page.

AMD.com

Run the installer.

Under Advanced Options, make sure to check Factory Reset / Clean Install (Redefinição de fábrica). This is critical to completely wipe the broken modern driver registry.

Complete the installation and restart your PC.

Step 3: Prevent Windows Update from Overwriting Your Driver
Windows Update will quickly try to force-install the broken newer driver. To block this:

Open the Windows Start Menu, search for “Change device installation settings” (Alterar configurações de instalação do dispositivo).

Open it and switch the selection to “No (your device might not work as expected)” (Não).

Save changes. (This keeps your system safe while allowing standard Windows security updates to flow normally).

Step 4: Bypass the D3D12RayTracing Assertion Crash
Because UE5 checks if your graphics driver is up-to-date upon launching with Ray Tracing enabled, opening your project with driver 24.6.1 will cause an immediate crash pointing to line 4979 in D3D12RayTracing.cpp.

To bypass this verification check:

Navigate to your project’s directory folder.

Open the Config folder and open DefaultEngine.ini with Notepad.

Press Ctrl + F and search for RayTracing.

If you find the line r.RayTracing=True or r.RayTracing=1, change the value to False or 0.

Hope this helps anyone struggling to push forward with their MetaHuman workflows!

[ASunshin]

I’m having the same problem. Updated and nothing fixed.

[Mobin0913191 (Mobin091319)]

Hello @3dcinetv, how have you solved the issue? I am having the same problem. Changing the frame rate for the capture data file does not solve it.

[ASunshin]

When I deleted 5.8 and reloaded it, the FAB and Quixel plugins were missing and not downloaded the second time. Tried everything and the plugin still won’t download.