Game Capture 0.2.65 — Windows

Game Capture now starts at 720p60 / 4 Mbps, with Auto hardware-first H.264, 128 kbps streaming audio, and no microphone selected. Existing saved settings are preserved.

• Lower-cost software and hardware encoder policies, zero B-frames, and approximately one-second keyframes.
• Baseline H.264 signaling/encoding alignment, including the AMF constrained_baseline setting.
• A 1,200-byte WebRTC codec-fragment limit for additional packet overhead headroom. SRT/UDP/RIST retain 1,316-byte payloads; SRT/RIST start with 200 ms recovery buffering.
• An illustrated encoder guide with bitrate/resolution recipes, hardware choices, measured resource costs, and WHIP cautions.

See the signed Windows qualification report for actual workflows, retained failures, focused reruns, hashes, and limits. These are Windows x64 downloads; the separate macOS beta is unchanged.

Known limits remain: occasional WHIP-to-RTSP relay-file startup timestamp duplication, receiver-dependent AV1/HEVC compatibility, and unqualified clean-machine/HDR/overnight scenarios. Signing identifies Social Stream Ninja; this test PC reports an untrusted certificate root, and signing does not guarantee SmartScreen reputation.

Setup, portable EXE, portable ZIP, stable download aliases, SHA-256 checksums, and FFmpeg license/source information are included. Built and qualified locally; no GitHub Actions build was requested.

macOS 0.2.65 Beta 3 — Experimental

Experimental Apple Silicon beta — Windows remains the production release.

Requires Apple Silicon and macOS 14.2 or newer. The app is Developer ID signed, notarized and stapled. Stop streaming and quit before replacing an earlier beta. App version remains 0.2.65; replace earlier 0.2.65 betas manually.

This candidate bounds Apple VT H.264 bursts while preserving average bitrate control and low bandwidth on simple scenes. Beta 2 reconnected with “WHIP media send failed” on a frozen complex-source transition. The new candidate preserves all frames in six 60-second landscape/portrait recordings; a constant-bitrate alternative also passed but used larger short bursts and more bandwidth on simple scenes. It retains Constrained Baseline, no B-frames, one-second keyframes, Auto → VT → x264 and the existing 720p60 / 4 Mbps window defaults. x264, HEVC, pixel conversion and Windows runtime are unchanged. No new VDO.Ninja API is introduced.

Runtime source: c7104c5cccce2878ded041489bf1285e409e9723.
Executable SHA-256: c0f3612991c1fb5bad48cd2ce999c889faf8c31acd0f9528dfed77beacda6c7a.
App archive SHA-256: aae737d3f63490280feea6054f0c2e345b08ef51b676d6bda4d1cc736e5b537b.
Complete source/build archive SHA-256: 7e9abd43808431540f3f5eeaf1567319f623c3e497a89bf0f710dab9671f9544.

Download the app ZIP, source archive and SHA256SUMS together; run shasum -a 256 -c SHA256SUMS. Corresponding AGPL source and pinned native dependency/build materials are included.

Actual signed-candidate testing passed native camera-to-WHIP recordings, existing VDO.Ninja live bitrate changes with invalid-token rejection, 1080p30 decoded output, stereo/silence, late viewers, GUI profile save/relaunch, receiver-server restart and Quit cleanup. Synthetic-source quality results slightly favor constant bitrate; the chosen bounded-average policy targets burst reliability and adaptive bandwidth, not a claimed quality increase. It is not a strict total-network bitrate ceiling.

The exact notarized ZIP passes browser download with quarantine preserved, settings/relaunch, thirty-minute 720p60 VT streaming with continuous stereo, a separate x264 media workflow and Beta 2 manual replacement with preferences retained. Fresh macOS 14.8.9 ARM64 VM startup/diagnostics/refusal workflows also pass (run). These do not qualify physical minimum-OS devices. The public GitHub ZIP also passes a real browser download with quarantine preserved, GUI settings/relaunch, and separate one-minute VT/x264 720p60 video plus stereo workflows with final controls and shutdown. Signing, source, metric and hash checks are separate gates. Physical minimum-OS/privacy/USB/sleep-wake/display/VoiceOver coverage, another Mac, real gameplay/thermal qualification and broader Windows hardware remain open. Keep Mac experimental.

Native benchmark evidence (repository access required).

Beta 3 validation and remaining limits.

macOS 0.2.65 Beta 2 — Experimental

Experimental Apple Silicon beta — Windows remains the production release.

Requires Apple Silicon and macOS 14.2 or newer. Extract the ZIP and move Game Capture.app to Applications. The app is Developer ID signed, notarized and stapled. Stop streaming and quit before replacing an earlier beta. The application version remains 0.2.65; Beta 1 users should replace the app manually.

This beta fixes Mac window-capture cursor inclusion to match Windows, preview oversaturation on wide-gamut displays, and source thumbnails stuck on fallback icons. Apple VideoToolbox hardware H.264 and x264 software H.264 remain the primary encoder choices. No new VDO.Ninja API or signaling command is introduced. Windows runtime changes are excluded.

Runtime source: ea1c552c6c833fecc16914e18710a589a6bb2db1.
Executable SHA-256: 2a6354f12b92ac1d8ef2a415c76316ea8452b040503087c1b8ec716be69ad1b0.
App archive SHA-256: f35f6852d8bc0b835211dc628fcd4d07b90e5580ccf7a64bb732fdead168b1db.
Complete source/build archive SHA-256: 4a614689d0df9eae9807dac69588b829c81dc11baac1e9f2992dd1dfbba0a1f2.

Download the app ZIP, source archive and SHA256SUMS together; run shasum -a 256 -c SHA256SUMS. The source archive contains application sources, pinned native dependencies and matching build materials.

Actual signed-package testing covers the GUI fixes with before/after screenshots, VT/x264 cursor exclusion through live enlargement and restoration (3,241 received frames), and native OBS H.264 color plus VP9 alpha composition through source aspect changes (2,636 frames). The delivered archive additionally passes local browser download with quarantine retained, GUI settings/relaunch, replacement of released 0.2.65 Beta 1 with preferences preserved, and separate 720p60/4 Mbps video/stereo workflows. Signing, notarization, source completeness and hashes are separate gates.

Fresh macOS ARM64 VM startup and x264 media workflows pass on the exact archives. These provide functional VM coverage, not physical hardware VT/audio or sustained 30 fps qualification. The public GitHub browser download also passes original-quarantine preservation, GUI launch/settings/relaunch and separate VT/x264 720p60 video/stereo workflows. Delivered Beta 2 results and limits. The released Beta 1 thirty-minute endurance results belong to its earlier executable; they do not automatically qualify this candidate. Physical minimum-OS/privacy/device/VoiceOver, sleep/wake, display/audio-route changes, broader Windows hardware coverage and matched OBS/FFmpeg/WHIP/gameplay quality remain open. Earlier unexplained VFR alpha pixels, hosted-motion, recorder-clock and abrupt-viewer-close findings remain recorded.

Candidate results and limits. Keep this Mac build experimental.

macOS 0.2.65 Beta 1 — Experimental

Experimental Apple Silicon beta — not the Windows production release.

Requires an Apple Silicon Mac and macOS 14.2 or newer. Extract the ZIP and move Game Capture.app to Applications. The app is Developer ID signed, notarized and stapled. Stop streaming and quit before replacing an earlier beta; saved-settings upgrade is covered by packaged application testing.

This beta incorporates the Windows 0.2.65 release baseline and the Mac capture, audio and publishing backends. Apple VideoToolbox hardware H.264 and x264 software H.264 are the primary encoder choices. It includes the persistent microphone-unavailable warning, Mac x264 presentation-time budgeting, native pixel conversion and the platform controls described in the architecture guide. No new VDO.Ninja API or signaling command was added.

Runtime source: de8a9410dc6f4ad9a7e65dd5ad5593e8ced74b15.
Executable SHA-256: 301dacfd38a6a4321ea548a68c674c499c7350fe651616968d020046a8f190aa.
App archive SHA-256: 06ce69dc8bcfe5e0406fc684893b2dca21c04a318f50b4b539bb5b039b916c38.
Complete corresponding source/build archive SHA-256: a08adf438d506adbaa2663411308480e5d4d693400487c5450841b5f5a88fefe.

Download the app ZIP, source archive and SHA256SUMS together; run shasum -a 256 -c SHA256SUMS. The source archive includes the app, pinned native dependency sources and build materials matched to the bundled binaries.

Actual testing on an Apple M1 covers VT/x264 microphone loss and recovery, live mute/unmute, No microphone restart, browser download with preserved quarantine, GUI settings/relaunch, 0.2.64-to-0.2.65 upgrade and separate 720p60/4 Mbps video/stereo workflows. Upgrading from public Beta 2 preserves settings; both encoders load its saved WHIP credentials without re-entry, publish authenticated video/stereo and recover after a local server restart. Fresh macOS 14.8.9 ARM64 startup and x264 media workflows pass on these exact archives. The VM delivers about 15–23 distinct images/s in moving/resize/late-viewer phases; this is functional coverage, not sustained 30 fps, hardware VT, audio or physical-device qualification. Windows packaged regression passes 70 GUI/shutdown assertions, 250,328 complete concurrent snapshot reads and WGC/software-H.264 moving/static/resize/late-viewer/source-loss workflows. The first Windows attempt correctly refused a stale QA version request before application testing; the corrected runner reads the CMake project version. Its QA-only changes do not alter the runtime source. Signing, source completeness and hashes are separate gates.

The minimum macOS 14.2 version, fresh physical-device/privacy/accessibility behavior, sleep/wake and audio/display-route changes remain unqualified. Earlier VFR artifacts retain unresolved hosted-motion, alpha-pixel, recorder-clock and abrupt-viewer-close findings; successful short workflows do not clear them or establish matched OBS/FFmpeg/gameplay quality. Prior binaries' endurance/latency results do not automatically qualify this executable. See the artifact-specific report for the recorded results and limits.

Windows stable remains the production release. This Mac beta remains experimental.

macOS 0.2.64 Beta 2 — Apple Silicon (Experimental)

Experimental Apple Silicon macOS beta. Windows v0.2.64 remains the stable production release.

Beta 2 fixes a room guest occupying a slot for roughly 90 seconds after its transport had already closed and its final BYE was not processed. Terminal guest connections now clear promptly; transient disconnections keep their recovery window. The change uses existing VDO.Ninja behavior and introduces no new API call, acknowledgement or relay. Unsupported native leave/unpublish sends were also removed.

Both encoders also passed fifteen-minute four-viewer room workflows, the existing connection-refresh path, sixteen total randomized controls, 256 local-control requests, and four GUI stop/start cycles. Local Beta 1 → Beta 2 replacement passed with seven private GUI preferences retained. Eight guest departures with their BYEs deliberately suppressed cleared in 76–329 ms through the terminal-transport cleanup path.

Packaged application testing on an M1 running macOS 26.4.1 covers VideoToolbox and x264: deliberately lost guest BYE, signaling recovery, public WHIP relay and receiver reconnection, audio-input and window-source recovery, and two-minute 1080p60 moving-detail capture with continuous received audio.

Built for macOS 14.2+, Apple Silicon only. Developer ID signed, notarized and stapled. Download the ZIP, extract it, and open Game Capture. Beta users should update manually; the stable update channel does not promote experimental prereleases. The app's internal version remains 0.2.64.

Production qualification is still incomplete: fresh/minimum-OS and second-Mac coverage, physical device and permission changes, sleep/wake, independent receiver and longer representative gameplay, trusted remote RTMPS, and a packaged Windows regression of the shared changes remain outstanding. The shorter local 1080p60 passes do not resolve the earlier three-minute workload failures. All current receivers run on the same Mac.

Matching source/build materials and SHA-256 checksums are attached. Runtime source revision: 1eefab24dbb03fa6bbe8caac30e2133865e71484.

Signed executable SHA-256: 2c1bfb2bc873c4c2c998034ec7ed5a311c2d0c3b99dd0d203305fef5ded1fad3.

The published ZIP also passed an installed-Chrome download and normal quarantined Launch Services workflows, including App Translocation, replacement/relaunch and private preference retention on this existing Mac. See the full qualification record and remaining requirements.

macOS 0.2.64 Beta 1 — Apple Silicon (Experimental)

Experimental Apple Silicon build. macOS is still beta; Windows v0.2.64 remains the production release.

Download game-capture-macos-arm64-0.2.64-beta.1.zip, extract it, and open Game Capture.app. The app is Developer ID signed and Apple notarized. Intel Macs are unsupported. The bundle targets macOS 14.2 or later; minimum-OS and fresh-machine qualification remain open.

This candidate carries fractional Mac audio timestamp durations between callbacks and integrates the current shared pacing/recovery code. The Windows audio-duration calculation is unchanged.

Actual signed-app workflows on one M1 Mac/macOS 26.4.1:

• GUI launch, diagnostics, invalid selections, clean shutdown, and VT/x264 720p60 window/application-audio streaming with live controls and continuous received-audio checks passed.
• The thirty-minute x264 paired recorded-media A/V comparison passed: −21.75 ms median drift within the unchanged 30 ms limit, with every event within the 50 ms skew limit. The previous beta failed this workflow. This improvement cannot be attributed to the fractional-duration change alone.
• VideoToolbox also passed its thirty-minute comparison: −14.5 ms median drift, with all paired events within 50 ms.
• All fourteen local VT/x264 output cases passed, as did the repeated four-viewer room workflow and eight additional VT guest cycles. One initial guest-cleanup check failed after rejoin; twelve later departures passed, but that intermittent failure remains unresolved.
• Local archive replacement from 0.2.62 to 0.2.64 passed, retaining one private UI preference and verifying both app identities.

A deliberate lost-bye control verified inactive guest tracks, cleanup after 90.25 seconds and continued media for the remaining viewers. It does not establish the cause of the earlier natural failure.

These controlled workflows do not establish production readiness. Fresh/minimum-OS and second-Mac coverage, physical devices and permission transitions, sleep/display/audio-device changes, remote services/trusted RTMPS, and integrated Windows regression remain open. 1080p60 production performance is unqualified; the previous beta missed sustained targets on the cohosted M1 under substantial remote-desktop load. Recorded-media A/V comparisons do not measure physical display/speaker or OBS presentation latency.

See the qualification record for measured scope, retained failures and remaining work.

The matching game-capture-macos-arm64-0.2.64-beta.1-source.tar.gz includes application and dependency source/build materials, notices and actual rebuild provenance. All application units were rebuilt from 8c34eb37fdad851e29e9577595af7a1f63711de4 using existing dependency libraries. SHA256SUMS.txt covers the two archives; the source archive’s manifest.json binds source payloads and all sixteen bundled binaries. The application is AGPL-3.0; component notices accompany the app and source archive.

Game Capture 0.2.64

Auto encoding now tries AMD AMF before software when Windows cannot provide a usable hardware Media Foundation encoder. Working hardware encoders retain their priority; explicit Software stays software. If AMD startup fails, Auto falls back safely. The streaming default remains 4 Mbps.

• Tested on RX 9060 XT: H.264 and AV1 AMD playback with audio and reconnects, plus VP9 software playback.
• Tested Auto choosing AMD and recovering to software when AMF is deliberately made unavailable.
• AMD H.264, HEVC and AV1 produced clean output in separate bundled-FFmpeg driver checks.
• Both 30-minute endurance runs passed: 319 dual-stream viewer cycles and 101 reconnect/playback iterations, with no retries. Packaged desktop, browser controls, updater failure handling and native OBS transparency workflows also passed; the validation notes retain initial runner failures and their focused reruns.

Codec compatibility: use H.264 with the tested OBS Ninja receiver. That receiver supports H.264/VP9, so HEVC playback remains unqualified there. AV1 and HEVC remain experimental VDO.Ninja options; external RTMP/SRT/WHIP/RTSP/UDP/RIST outputs remain H.264. VP9 encoding is software.

Windows package validation · AMD codec results · Windows guide

Choose game-capture-setup.exe, game-capture-portable.exe or game-capture-win64.zip. SHA256SUMS.txt and FFmpeg source/build information accompany the downloads. Application/setup are signed by Social Stream Ninja; this test host reports an untrusted certificate root. Existing WHIP loss/jitter and receiver startup-timestamp limitations remain documented. The macOS beta is unchanged.

Game Capture macOS 0.2.62 Beta 1 — Experimental (Apple Silicon)

Experimental beta for Apple Silicon Macs only. This is an early Mac port, not a production-qualified Mac release. The tested Windows production release remains Game Capture 0.2.64.

Download and install

Download game-capture-macos-arm64-0.2.62-beta.1.zip, extract it, and move Game Capture.app to Applications. Launch the app and grant screen recording, camera, or microphone permissions as needed for your selected sources.

• Apple Silicon (M-series / arm64) only; no Intel build.
• Built for macOS 14.2 or later. Packaged application testing was performed on an M1 running macOS 26.4.1; the minimum OS and clean-machine installation are not yet qualified.
• Apple VideoToolbox hardware H.264 and x264 software H.264 encoding.
• Developer ID signed, accepted by Apple notarization, with a stapled ticket. Gatekeeper assessment passed locally.
• The app's internal version is 0.2.62. Beta 1 identifies this separate experimental Mac distribution.

Packaged application testing

The exact signed ZIP was extracted and exercised through application workflows: GUI launch and saved settings; window publishing with selected-application audio using VideoToolbox and x264; and controlled virtual-camera/audio-input WHIP publishing with both encoders. Receiver video/audio output and lifecycle behavior were checked. These results do not establish full Windows feature parity or broad Mac compatibility.

Known limitations

• A/V timing blocker: the long-run VT recorded-media comparison passed, but x264 failed: median offset changed by −44.75 ms against a 30 ms drift limit, and some final events exceeded the 50 ms skew limit. A Mac-only fractional-timestamp correction is under development; it is not in this download, does not explain the complete measured drift, and has not passed packaged application testing. The release remains experimental.
• Endurance follow-up (October 16): the unchanged signed beta now passes complete thirty-minute VT and x264 window/application-audio workflows, including continuous received-audio observation and subsequent functional controls. Saved pixels from an x264 identity rejection identified cursor interference with the observer; the corrected reader passed an occlusion control and still rejected a deliberate frame reversal. The earlier VT rejection lacks saved pixels, and a separate brief audio-concealment transient remains unexplained. Supplementary browser recordings have timestamp gaps and are not a gapless timing reference. See the Mac testing status and open findings for exact scope and remaining production requirements.
• RTMP audio update (October 16): the earlier missing-audio result was traced to the FFmpeg 7.1.1 RTSP recording tool dropping received AAC packets during stream-copy. The unchanged signed beta now passes local RTMP video/stereo-audio, silence/resume, and reconnect/late-receiver workflows with VideoToolbox and x264, through both direct RTMP and corrected RTSP reception. FFmpeg 9 RTSP reception also retained audio in the comparison. This resolves that specific failure; remote services, RTMPS, and broader receiver compatibility still require qualification.
• A real Chrome download retained quarantine and passed normal macOS Open prompts, App Translocation, GUI controls, preference retention during same-version replacement, and clean shutdown on this existing Mac. Clean-machine permissions, minimum-macOS and second-machine compatibility, cross-version upgrades, physical devices and sleep/display changes remain incomplete. Broader workload/network coverage remains incomplete.
• The port's shared-code changes have not completed Windows regression qualification and have not been merged into the Windows production release.

Please report Mac issues with your macOS version, Mac model, source type, encoder, and output destination. Remove stream keys, passwords, and private URLs from logs before sharing them.

Source and integrity

game-capture-macos-arm64-0.2.62-beta.1-source.tar.gz contains the matching application source, bundled dependency source materials, patches, notices, and build instructions. Use this asset when rebuilding the distributed app; GitHub's automatic source ZIP/tarball contains the repository snapshot only, without the bundled dependency source materials. The application is AGPL-3.0; component notices are included in the app and source archive.

The release tag points to the exact repository snapshot included in the source asset. Download filenames include the beta label; the signed app and source archive bytes are unchanged from the locally verified candidate. SHA256SUMS.txt covers both downloadable archives.

Game Capture 0.2.63

Game Capture 0.2.63 adds custom resolutions and the recent streaming reliability improvements.

• Type custom even resolutions in Advanced settings, with 1440p, 4K and portrait presets. Valid custom sizes survive restart; invalid sizes are rejected before publishing.
• RTSP over TCP/UDP, UDP MPEG-TS and RIST publishing join VDO.Ninja, WHIP, SRT and RTMP/RTMPS outputs.
• New configurations use 4 Mbps and hardware-first encoding. Software fallback, encoder pacing and high-resolution capture handling reduce unnecessary load. Existing saved preferences are retained.
• Microphones are opt-in, with a None choice and mute control. Missing audio devices no longer stop otherwise healthy video publishing.
• Audio-format fallback, fractional sample-period handling, AMD encoder startup/header corrections and bounded WHIP NACK/RTX recovery include the latest fixes.
• VP9 alpha output now limits cold-start quantization to prevent a first-frame color shift, verified through repeated OBS playback.

Known limitations: Chromium WHIP playback can still pause under packet loss/jitter; disabling RED or increasing buffering is not a demonstrated fix. MediaMTX-to-FFmpeg RTSP reception can produce duplicate timestamps near startup. Physical AMD hardware/driver qualification and clean-machine testing remain outstanding.

Download the setup, portable EXE or portable ZIP below. Stable filenames remain available for existing download links. FFmpeg source/build information and SHA256SUMS.txt are included.

The application and installer carry a Social Stream Ninja signature, whose root is not publicly trusted on the validation host.

See the release qualification report and detailed recovery/performance results.

Game Capture 0.2.62

Audio controls and additional outputs stay tucked into Show advanced settings. The default remains VDO.Ninja with 48 kHz stereo Opus at 192 kbps.

• Set Opus bitrate and mono/stereo output. PCM and Opus RED are experimental options for compatible VDO.Ninja receivers; PCM+RED and surround output are not enabled.
• Publish H.264 directly to WHIP, SRT, RTMP or RTMPS. Destination profiles remember their settings and encrypt saved credentials for the current Windows user.
• WHIP supports endpoint-provided TURN servers, enforced Relay Only operation, secure session cleanup and faster recovery when receiver feedback stops.
• Capture cadence, media timestamps, audio transitions and resource cleanup include the quality fixes developed alongside these outputs.

Packaged validation includes real Chrome playback, Meshcast, remote Windows/macOS Chrome sessions, lossy SRT, trusted TLS destinations and recorded OBS output. A 35-minute native OBS recording passed the audio continuity indicators and measured A/V skew stayed within 20 ms in beginning/middle/end samples. Further soaks were explicitly skipped. These are measured results for the documented setups, not a guarantee for every receiver.

OBS browser-source limitation: OBS 32.2.2 reproduced an audio discontinuity after silence even with audio generated entirely inside OBS, without Game Capture. A native OBS Media Source using the receiving server's RTSP URL is the tested alternative. Browser-source video delay may also need calibration. Experimental PCM requires a compatible Chrome viewer; Firefox users should select Opus.

See the qualification report and output setup instructions for coverage, measurements and limitations. Asset checksums are in SHA256SUMS.txt.

The OBS image checker was calibrated for measured VP9 startup quantization, with its existing defect controls preserved. Original failed runs and the final calibrated playback results are retained in the qualification report.