Give each moment context
Review action, humor, commentary, lore and quieter discoveries with cited observations and setup/payoff timing. Completed reviews are reused; first-time analysis takes longer.
UPDATES / CHANGE LOG
What changed in ReplayFoundry.
New features, fixes, and earlier releases—all in one place.
CURRENT DOWNLOAD
1.0.0 BETA 5.9
RELEASED
SEPTEMBER 27, 2026
1.0.0 BETA 5.9
RELEASED
SEPTEMBER 27, 2026
Available now
A clearer workspace. Carefully reviewed learning.
Beta 5.9 simplifies Settings, smooths editing interactions and adds reviewed community-writer releases. The existing writer stays active until a real candidate qualifies.
These improvements are included in the current signed 1.0.0 Beta 5.9 download.
CURRENT RELEASE / HIGHLIGHTS
What’s new in this beta.
Listen across the tracks
Separate speech tracks and local acoustic clues add context. Unknown or mixed audio stays uncertain, so a game character is not automatically treated as the creator.
Write from supported evidence
Titles and descriptions draw on supported moments and attributed speech from the actual cut. Every draft stays editable and needs your review.
Keep Pop captions word by word
Preview and export share word timing. Review flags take you to unresolved words, overlaps and possible pauses; search and replace help with text corrections.
Say what missed the mark
Rate a clip below playback and optionally flag a wrong category, speaker, missing setup or payoff. These corrections stay separate from taste training.
Know what you have shared
See clips, a month calendar and the selected day’s agenda together. Shared publication states show what is scheduled or published, with an explicit action for reposts.
PUBLIC RELEASES
The record so far.
Release notes describe what was actually available on the date shown.
Read the technical change log01
1.0.0 Beta 5.9
Clearer Settings, smoother editing and reviewed community-writer release support.
- Find settings in five clear groups with a compact, wider search field.
- Keep creator-voice drafts when switching sections. Caption controls, preview timers, nested scrolling, Library seeking, and Publish dragging respond with less redundant UI work.
- Add reviewed community-writer evaluation and signed-release support. Training alone cannot publish a model; factual, privacy, wording, and timing checks must pass before owner approval.
- Keep text sharing optional and off by default. Training examples, private review records, and release credentials are excluded from distributed model packages.
- Update Advanced AI to runtime 0.8.35 and model package 4.0.31. Pretrained weights are unchanged. No community-trained model is activated in this release.
- Existing projects and feedback are retained. Review generated wording and captions before publishing.
02
1.0.0 Beta 5.8.1
Longer wording edits and clearer optional text sharing.
- Retain saved descriptions up to 5,000 characters and edits that remove every tag, with exact additions and removals linked to the original scene evidence. The bounded record size still applies.
- Keep extended edits in training exports with a clear exclusion reason when they cannot fit the original generation contract. They are not silently discarded or forced into model training.
- Read an accurate Privacy summary for the separate, default-off writer-sharing opt-in. It shares future text and metadata only; no audio, video or images are uploaded. Stop-sharing, retry and deletion controls remain available.
- Includes Beta 5.8 writing controls: separate Rewrite and New angle actions, saved defaults, field locks and protection for newer typing while a reroll finishes. Publish keeps a readable full-height activity column.
- Update Advanced AI for runtime 0.8.34 and model package 4.0.30. Pretrained weights are unchanged, DeepSeek remains experimental, and no new personal model is activated.
- Existing projects and feedback are retained. Review generated titles, descriptions and captions before publishing.
03
1.0.0 Beta 5.8
Evidence-bound writing feedback, clearer rewrite controls, opt-in text contributions and a readable Publish activity column.
- Save exact title and description additions and removals with their original scene evidence for future writer learning. Unchanged fields do not become approval labels.
- Use separate Rewrite and New angle actions, saved writing defaults and field locks. Newer typing is preserved when a pending reroll finishes.
- Keep clips, calendar, uploads and recent activity together, with a readable full-height activity column and room for upload progress.
- Choose whether to help improve the writer in Settings → Privacy. The separate default-off opt-in shares future text and metadata only, with stop-sharing, retry and deletion controls. No audio, video or images are uploaded.
- Update the app from Settings, then update Advanced AI for runtime 0.8.33 and model package 4.0.29. Pretrained weights and the default writer are unchanged; DeepSeek remains an experimental candidate.
- Saved projects and feedback are retained. No new personal model is activated by this release. Review generated titles, descriptions and captions before publishing.
04
1.0.0 Beta 5.7
A unified Publish workspace with clips, a month calendar and daily agenda.
- Publish from one workspace with your clips, month calendar, daily agenda, upload progress and recent activity together.
- Browse a compact clip list that scrolls independently. Select a clip to prepare it, open its existing YouTube upload, or explicitly prepare a repost.
- See small status icons and counts on calendar days, then select a day for full titles and times. The selected day stays in place when the schedule refreshes.
- Keep search, date, folder and publication filters close to the clips. Connection options, full history and performance observations remain available without stretching the page.
- Update the app from Settings. The verified Beta 5.6 Advanced AI components are reused, so a current AI installation needs no new model download.
- Saved projects, feedback, montage styles and writing controls are retained. Review generated captions, titles and descriptions before sharing.
05
1.0.0 Beta 5.6
Clearer creation and publishing, editable montage styles, and stronger writing evidence.
- See shared YouTube publication status in Library, Publish and the calendar. Open existing uploads, distinguish public from scheduled videos, and prepare intentional reposts explicitly.
- Keep Library filters visible, group videos by game or project, and remember your view. Generate now uses three stages with your moment objective and a final review together.
- Choose Impact, Banter or Tension for montage pacing and audio joins. Save the editable sequence and its own title and description; changes to the cuts invalidate old sequence wording.
- Choose Angle and Tone beside the writing controls, lock either field, and keep saved wording when a reroll cannot pass review. Independently cited source text helps preserve dialogue roles and unfinished actions.
- Understand general AI, personal clip taste and personal writing separately. Explicit feedback and recording-separated quality checks govern activation; this update does not activate an underqualified personal model.
- Update from Settings, then update Advanced AI for the matching components. Saved projects and feedback stay on this PC. Pretrained weights are unchanged; first-time analysis still takes time.
- Review generated wording before sharing. Music beat editing, mixed-audio separation and broader semantic montage planning remain future work.
06
1.0.0 Beta 5.5
More reliable long-recording speech analysis with the same compatible AI packages.
- Long recordings keep their valid speech analysis when Whisper adds a tiny segment after an audio chunk ends. Retained speech and timestamps stay unchanged.
- Includes Beta 5.4 improvements: reuse completed recording analysis, review distinct alternatives in smaller groups, and ask for stronger evidence behind AI wording.
- Use Check for updates in Settings. The existing Beta 5.4 AI packages remain compatible, so they do not need another download for this fix.
- Saved projects and feedback remain on this computer. This update does not change model weights or train your personal model.
- First-time analysis still takes time. Review generated captions, titles and descriptions before publishing.
07
1.0.0 Beta 5.4
Less repeated processing, more selective follow-up review, and stronger evidence for AI wording.
- Changing clip preferences reuses the recording map. Resumed scans decode unfinished sections, and picture checks reuse identical visual work.
- Additional review checks two distinct moments at a time and stops when enough clips qualify. Routine interface work waits behind stronger moments unless it supports a joke, insight or reveal.
- Scene checks require evidence for individual claims and numeric details. Dialogue attribution can still be wrong, so review each title and description before sharing.
- See an estimate for the current recording scan. During additional review, Finish with ready clips lets you complete captions, writing and saving for the accepted clips already available.
- Use Check for updates in Settings, then Update Advanced AI for the matching components. Your saved projects and feedback stay on this computer.
- First-time analysis still takes time. Pretrained model weights are unchanged, and this update does not train or activate a personal model.
08
1.0.0 Beta 5.3
Recovery for long scans and rejected cuts, confirmed game names, and more reliable Pop caption repair.
- Long recording scans keep completed sections when their time budget expires. Recovery checks the recording and analysis identity before reusing saved work.
- Automatic cuts need picture review covering the full clip. If rejected cuts leave too few results, ReplayFoundry checks more alternatives within the existing review limit.
- A cut that cannot support reliable AI wording can be replaced while eligible alternatives remain. Rejected cuts stay excluded, and generation does not silently switch to simple titles.
- A game name you confirm for a recording is preserved when visual analysis mistakes a tutorial heading for another title.
- Pop caption repair fills missing word timestamps while keeping usable existing timing. Newly aligned words need a sufficient acoustic match and cannot overlap retained words; unresolved timing still requires review.
- Install over your existing app or use Check for updates in Settings. Update Advanced AI installs the matching runtime packages. Saved projects and feedback stay on this computer.
- AI picks and writing still need your review. This release updates recovery and validation; pretrained model weights are unchanged and no personal model is activated automatically.
09
1.0.0 Beta 5.2
Audio-informed moment review and writing, caption timing review, Studio search, feedback reasons, and organized export files.
- Moment review now considers action, humor, commentary, lore, discovery, failure, clutch, tutorial and reaction, with cited timing and setup/payoff context.
- Separate audio tracks and a local acoustic model add speech and gameplay clues. Unknown or mixed tracks remain uncertain; track names alone do not prove who spoke.
- Supported scene and speech evidence also inform title and description drafts. Initial analysis takes longer, completed evidence is reused, and drafts still need your review.
- Pop captions use individual word timing in preview and export. Unresolved timestamps require review; caption search, replacement and timing navigation make corrections easier.
- Find Studio clips by title, description, recording or captions. Exports put subtitles, thumbnails, metadata and editing handoffs in a Supporting files folder beside the finished videos.
- Optional reasons below the preview let you flag a wrong category, speaker or cut boundary. Factual corrections stay separate from personal taste training.
- Update from Settings: About & updates checks for signed releases, downloads the next version, and restarts when your work is ready. Automatic checks are optional. Existing Beta 5.1 users can get Beta 5.2 from Check for updates. The matching AI tools are available separately through Update Advanced AI in Settings.
- Quieter video navigation, smoother title and description typing, more forgiving game-name searches, and a fix for automatically loading duplicate caption sidecars.
- Writing choices are available before you edit. Clearer expandable sections and simpler learning settings make Studio easier to navigate.
- Balanced AI reviews the recording and checks a shortlist visually before choosing automatic clips. Completed rejections stay excluded from replacement choices; your marked ranges remain yours to choose.
- Titles and descriptions refresh stale scene evidence after edits. Remembered folder game names require confirmation for each recording, helping prevent unrelated game tags.
- If an automatic cut cannot support accepted AI wording, generation can try another already-reviewed moment. It does not silently switch to simple titles.
- Project Clips is wider, collapsible sections look different from dropdown fields, and Like / Neutral / Dislike sit below the preview playback controls.
- First-time visual analysis takes longer, while completed reviews are reused. AI copy can still be vague or inaccurate, and caption timing can need editing; review each clip before sharing.
- Wording corrections can explain what needs fixing, such as a wrong speaker or something that did not happen. Changing only a title no longer treats the unchanged description as approved.
- Local learning keeps references to the original clip and the order of exported montages for later review. Recordings reserved for testing remain separate as you add footage.
- Added lore/story moment labels and montage-specific writing review. Personal writer updates must pass independent checks before use; ordinary editing continues while you collect examples.
- Faster AI title and description writing, with shorter prompts and less repeated analysis when you request another version.
- AI stays ready during a writing session and releases its GPU memory when the session ends, is cancelled, or becomes inactive.
- Optional wording learning saves your edits and explicit approvals locally. Advanced AI includes the base model for future personal-writer training; a personal writer needs real feedback and independent quality checks before activation.
- Automatic gameplay picks screen out more launcher, game setup, and loading footage, including replacement choices.
- Studio uses more of your screen, and Library shows the same clip titles you edit and publish.
- Faster activity scoring and fewer background checks on unused Find More alternatives.
- Corrected Beta 5 installer: restore missing video tools and repair damaged local AI packages. Repair remains available if the saved installer is missing.
- Fixed a crash when clearing recent projects after deleting recordings. Calendar dates from neighboring months now have yellow numbers and month labels.
- Caption regeneration and word edits now work correctly after moving a clip to another part of the recording.
- Simpler menus, clearer wording, and a more consistent look throughout the app.
- Full-recording timelines with thumbnails, suggested moments, content filters, draggable clip ranges, and editing shortcuts.
- Short caption phrases that follow speech, with animated effect previews and easier styling controls.
- Title and description drafts for clips you find yourself as well as clips selected by AI.
- A personal neural model that learns locally from your clip choices, with quality checks before personal ranking activates.
- An updated local AI runtime, stronger privacy checks for support reports, and more reliable moment selection.
10
1.0.0 Beta 5.1
Signed in-app updates, quieter video navigation, smoother writing, easier game lookup, and clearer Studio controls.
- Added Check for updates and optional automatic checks in Settings.
- Reduced preview audio bursts and unnecessary refreshes while typing.
- Improved game lookup, Studio section styling and access to writing choices.
- Separated caption sidecars to prevent players automatically displaying duplicate subtitles.
11
1.0.0 Beta 4
A clearer workspace with more control over captions, framing, sound, writing, and delivery.
- A clearer workspace across Generate, Studio, Library, Publish, and Settings, including smaller windows.
- Reusable caption looks, editable words and timing, safe margins, and subtitle files for clean-video delivery.
- Portrait, square, and landscape edits with gameplay and facecam framing, crop waypoints, audio controls, and output through 2160p.
- Title and description drafts balance visible events with spoken commentary, retain earlier copy, and show findings that need review.
- Shared video analysis, responsive previews, rendering progress, and verified completed-clip recovery reduce repeated work.
- Local publishing packages and OpenTimelineIO handoffs carry the clip, subtitles, and metadata into the next editing or publishing step.
12
1.0.0 Beta 3
A more reliable creative pass with stronger moment variety, closer captions, more natural writing, and clearer feedback from setup through publishing.
- Finds a broader mix of moments, including action and visual payoffs alongside conversation.
- Keeps punchy and karaoke captions closer to each spoken word and natural pause.
- Writes more natural, distinct titles and descriptions from speech, visuals, and game context.
- Lets alternate moments prepare in the background and arrive in Studio with the same caption and writing pass.
- Adds clearer progress, playback, restored-thumbnail, scheduling, and YouTube history feedback.
- Refreshes setup and makes the optional Advanced AI choice and readiness state easier to understand.
13
1.0.0 Beta 2
The first Microsoft-signed public beta with an optional local AI installation path.
- Introduced a small signed Base installer with an optional Advanced AI choice during setup.
- Added verified local transcription, visual review, and grounded title and description rewrites.
- Refined Studio captions, Library organization, Publish scheduling, and interface feedback.
- Activated compatible runtime packs as one set instead of stacking old model copies.
14
1.0.0 Beta 1
The initial private Base build used to validate clean installation, the core four-workspace flow, and complete removal on a fresh PC.
- Included Generate, Studio, Library, Publish, and Settings without the optional local AI package.
- Focused on clean installation, core editing, local media tools, and uninstall behavior.
- Was unsigned and intended only for the first controlled installation tests.
CURRENT SIGNED BETA