[Bug] WebGLRenderingContext returns the final composited frame with the CEF/NUI layer

[zoovdev]

Summary

Capturing the game frame through a WebGLRenderingContext returns the final composited frame with the CEF/NUI layer in it, instead of the game render on its own like Legacy. This breaks screenshots (UI shows up in the image) and live in-world previews that draw the frame back into NUI, which feed back into themselves.

Issue type

Client

Repro rate

Always

Server build version

b94 win32

OS

Windows 11

CPU

AMD Ryzen 9 9950X3D2

GPU

NVIDIA RTX 5090

RAM

96 GB DDR5

Storage type

SSD NVMe

Connection type

Ethernet

ISP and bandwidth

1000/500

DxDiag

No response

Network graph

No response

Platform

None

OS version / distribution

Windows 11

CPU

Ryzen 9 9950X3D

RAM

96 GB DDR5

Using txAdmin?

Yes

Hosting provider

Self-hosted

Machine type

Dedicated

/perf endpoint output

No response

DDoS protection

No

ulimit -n value (Linux only)

No response

Docker Compose file (Docker only)

No response

Steps to Reproduce

Edit: DLSS must be enabled to replicate, turns out this is only occuring, along with some infrequent crashes or black screens while DLSS / frame generation is enabled.

1. Start any resource that draws an NUI layer on screen (a menu, overlay, HUD, inventory, anything visible in CEF).

2. Capture the in-game frame with the WebGL gameview method. Bind the game framebuffer to a WebGLRenderingContext texture, draw it to a canvas, then read it back with drawImage or toDataURL, or just display it live. The public screencapture and screenshot-basic resources both use this path, so exports.screencapture:requestScreenshot(...) is enough to reproduce it.

3. Save or preview the result. The NUI layer is baked into the captured frame.

It happens every time. If you sample the framebuffer every frame and draw it back into an on-screen NUI element (for example a 3D weapon or vehicle preview pane), you get a feedback loop. The UI gets drawn into the preview, that preview is part of the next captured frame, and it repeats until the whole thing is nested boxes.

Expected Behavior

The WebGL capture of the game framebuffer should only contain the game render, before the NUI is composited on top. On Legacy this is how it works. Drawing the game framebuffer to a canvas through a WebGLRenderingContext and reading it back gives you the 3D scene with no UI painted over it.

Ideally this would be selectable instead of one fixed behavior. Both frames exist, the game render before the UI and the final composited frame, so it would be useful to pick which one the WebGL capture returns.

Actual Behavior

On GTA V Enhanced the framebuffer you get in the WebGLRenderingContext is the final composited frame, so it already has the CEF/NUI layer in it. Screenshots taken this way show the UI in the image, and any live preview that draws the captured frame back into NUI feeds back on itself and is unusable.

Evidence

No response

Additional Context

https://github.com/itschip/screencapture

[zoovdev]

Additional step for replication: DLSS must be enabled.

When disabling DLSS, it works as intended.