LuminalShine 26.08.1

LuminalShine 26.08.1 is a stability-focused release for the LuminalVGD direct-to-encoder path introduced in 26.08.0. It addresses the shared GPU-resource lifetime and recovery failures that could produce a frozen picture, brief continued audio, and eventually an unreachable host. It also corrects worker startup, requested display-mode handling, topology cleanup, and interrupted state-file writes.

This release continues to bundle the signed LuminalVGD v0.1.0-alpha.7 / Build 23 package used by 26.08.0:

• Driver version: 0.1.0.23
• Driver build: 23
• Protocol: 0.9
• Package SHA-256: afe890d0b0b94232b20eb0b6e3aac738e2f1ea73c66572541b9bb1b2e97d5afe

The release build downloads the signed driver archive from the LuminalVGD GitHub release and verifies it against the pinned SHA-256 digest. Signed driver binaries and catalogs are not stored in the LuminalShine source repository.

Recommended update: Hosts running 26.08.0 should upgrade to 26.08.1, especially systems using LuminalVGD direct capture, HDR, high refresh rates, HEVC 4:4:4, or long-running sessions.

Highlights

Direct capture is decoupled from encoder lifetime

LuminalVGD ring textures are no longer kept in an encoder-owned lifetime while NVENC or another encoder is consuming a frame.

• Claimed LuminalVGD frames are copied into LuminalShine-owned mailbox textures.
• The driver ring slot is released after the bounded copy instead of remaining coupled to encoder progress.
• A stalled or rebuilding encoder cannot indefinitely stop the driver producer from recycling its ring surfaces.
• Ring ownership, generation, transport, and slot validation remain enforced before a frame is accepted.
• Direct capture retains the negotiated resolution, refresh rate, HDR state, and chroma behavior.

This changes the failure boundary: encoder latency or teardown can no longer hold LuminalVGD's IddCx-facing surfaces hostage and turn a local encode problem into a display-wide resource deadlock.

Capture and encoding worker isolation

The latency-sensitive capture and encode pipeline now runs in a bounded worker process for supported streaming sessions.

• The main LuminalShine service supervises the worker with startup, readiness, liveness, and shutdown deadlines.
• A capture or encoder failure is contained to the worker instead of permanently wedging the host control plane.
• Worker readiness requires a real first encoded packet, not merely successful process creation.
• The worker receives the encoder capabilities already validated by the parent, avoiding a second full encoder probe during client connection.
• Startup errors and first-frame failures are returned explicitly to the parent for deterministic cleanup.

Exact LuminalVGD session handoff

Process isolation initially exposed a process-local state problem: a child process could not infer the parent's active virtual-display session and could fall back to desktop capture. The handoff is now explicit and fail-closed.

• The parent transfers the exact active LuminalVGD session and display identity.
• Ring generation, slot count, selected transport, dimensions, format, and active state are validated in the worker.
• Stale or mismatched ring identities are rejected.
• A direct LuminalVGD worker cannot silently substitute Windows Graphics Capture or Desktop Duplication.
• Authentication and process identity checks protect the local worker-control channel.

Requested client display modes are authoritative

Direct LuminalVGD sessions now preserve the client-requested mode throughout display preparation and worker startup.

• Client resolution and refresh-rate requests are enforced for direct capture.
• Capture readiness is tied to the requested virtual display rather than whichever physical display Windows currently exposes as primary.
• Active-ring identity and dimensions must agree with the prepared display mode.
• Startup failures clean up the staged virtual-display session instead of leaving a partial topology behind.

This is particularly important for 4K120, HDR, and high-refresh clients whose requested mode differs from the local physical monitor.

Recovery and shutdown improvements

• Display-helper teardown uses phase-specific deadlines so a slow recovery step is not mistaken for a fatal global hang.
• Worker failure is isolated from the RTSP control plane, allowing the host to remain reachable for cleanup and subsequent connections.
• Direct capture fails closed when its LuminalVGD target is stale instead of entering an unintended WGC path.
• Virtual-display staging and cleanup retain enough identity information to remove the correct session after partial startup.
• GPU and ring recovery diagnostics now distinguish worker startup, first-frame, liveness, transport, and display-preparation failures.
• Physical-display topology restoration is separated from encoder shutdown, reducing cross-component teardown races.

State-file reliability

Atomic state persistence and recovery have been hardened for interrupted and concurrent writes.

• Unique temporary files are used instead of relying on a single shared .tmp pathname.
• Legacy temporary files and backups remain recoverable.
• A valid temporary or backup record can restore state after an interrupted replacement.
• Failed replacements retain recoverable data instead of silently discarding the last usable state.
• Cleanup recognizes the state files LuminalShine creates when running from the installed service context.

These changes cover display restoration and other persistent state used during launch, shutdown, and recovery.

Security and correctness

• Worker IPC uses constrained local access and verifies the expected peer process.
• Message sizes and transferred ring metadata are validated before use.
• The worker binds to the parent's exact LuminalVGD target and rejects stale generations or incompatible transport state.
• Encoder capability transfer is bounded and validated rather than reconstructed from untrusted or incomplete child state.
• Failure paths prefer an explicit connection error and cleanup over silently capturing the wrong display.

Upgrade notes

1. Download and run LuminalShine_x64-installer.exe from this release.
2. Existing paired clients and ordinary LuminalShine configuration are retained.
3. The bundled LuminalVGD driver remains Build 23 / protocol 0.9; a driver re-sign or driver-version change is not required for this host update.
4. Reboot only if Windows or the installer reports that an in-use component could not be replaced.
5. After installation, verify LuminalShine 26.08.1 on the About page.

Portable-package users should stop all LuminalShine processes before replacing files. Driver installation and updates remain privileged system operations and are best handled by the installer.

Validation

• Windows UCRT production build completed for luminalshine.exe.
• The C++ test binary built successfully.
• 67 focused tests covering LuminalVGD ring liveness, worker targeting, display-helper behavior, requested modes, state storage, and session identity passed locally.
• GitHub Windows AMD64 build and coverage completed successfully.
• CodeQL analysis passed for C/C++, Actions, C#, JavaScript/TypeScript, Python, and Ruby.
• Vitest coverage, Codecov patch analysis, and bundle-size analysis passed.
• The release source contains no signed driver packages, certificates, logs, ETLs, or locally generated binaries.

Known limitations

• A genuine GPU device removal, reset, or hang still destroys the active graphics-device generation. LuminalShine can contain the worker failure and preserve service reachability, but Windows and the GPU driver must successfully recreate the graphics device before streaming can resume.
• A machine-wide display-stack failure where Windows permanently rejects display-configuration queries can still require a reboot.
• Clients may need to reconnect after a sufficiently long GPU outage even when LuminalShine remains reachable.
• LuminalVGD direct capture deliberately fails closed if the worker cannot prove that it opened the exact prepared ring. It will not silently reduce fidelity or capture a physical display in place of the requested virtual display.

Downloads

• LuminalShine_x64-installer.exe — recommended installer, application update, and LuminalVGD deployment path
• LuminalShine_x64-portable.zip — portable application package
• LuminalShine_x64-debuginfo.7z — symbols for crash dumps and field diagnostics

The raw MSI remains available inside the GitHub Actions build artifact for internal diagnosis but is intentionally not attached to the public release.

Reporting problems

For frozen video, missing first video, incorrect virtual-display modes, or recovery failures, include:

• the complete LuminalShine runtime log, including worker-process output
• LuminalVGD, IddCx, and DDisplay ETLs covering the failure when available
• the client resolution, refresh rate, codec, chroma mode, and HDR state
• the selected virtual-display mode and layout
• relevant Windows events from Display, DxgKrnl, the GPU vendor driver, and WHEA

These details distinguish an encoder stall, worker startup failure, stale ring handoff, driver transport failure, client disconnect, GPU device removal, and machine-wide Windows display-stack failure.

Changes since 26.08.0

• Stabilize LuminalVGD direct capture and recovery
• Full source comparison