Skip to content

Releases

Cutline v0.8.8 — Color your canvas. Shape your captions.

Choose a tag to compare

@ahpah-dev ahpah-dev released this 09 Oct 18:02

Cutline 0.8.8 — Color your canvas. Shape your captions.

Two editing updates in one free, watermark-free Windows release.

Customizable Whisper caption line length

  • In Captions → Auto subtitles, choose Words per line from 1–20. Quick presets: 1, 3, 5 and 7 words.
  • Your choice is remembered locally and shown in a live example. Existing captions are untouched; new captions remain normal editable text clips.
  • Real Tiny word timings are retained. Large's multiword segments are split into shorter cards with estimated within-segment timing. Speech pauses can produce fewer words than the selected maximum.
  • Text-only transcription fallback now retains the decoded audio duration after its sample buffer is transferred to the worker. Consecutive caption cards do not retain overlapping minimum display tails.
  • Generation remains local, with the same Tiny/Large model choices. No new model download or paid transcription API is required by this setting. Review recognition and timing before export.

Gradient canvas backgrounds

  • Deselect the current layer and open Inspector → Canvas background. Choose Solid, Linear or Radial.
  • Six palettes; 2–8 exact hex colors; editable stop positions; add/remove/reverse controls.
  • Linear angle, radial center and radius, keyboard-accessible controls, normal Undo/Redo.
  • Included in project backups, autosave, AI previews and offline MP4/WebM exports. Old projects retain their solid backgrounds; switching to Solid preserves the gradient settings.
  • Gradients render behind layers, letterboxing, transparency and timeline gaps. Empty-player hints no longer dim their preview. A background does not add timeline duration on its own.
  • AI tools can inspect and precisely edit backgroundFill with the same strict, atomic revision checks.

Verification and downloads

Unit/component and native protocol tests, real rendering/export regression checks, and isolated Windows UI flows cover both updates. The caption UI flow uses a deterministic transcription-worker fixture with real local audio decoding; it does not claim a new accuracy benchmark for the unchanged Whisper models.

Use Cutline-Setup-0.8.8-x64.exe to install/update, or Cutline-Portable-0.8.8-x64.exe for the portable build. Verify downloads with SHA256SUMS.txt. Windows x64; the installers are unsigned and may trigger SmartScreen. Existing projects and model caches are not removed by this release.

Cutline v0.8.7 — Your models, your edit

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 21:40

Cutline 0.8.7 — Your models, your edit

  • Custom AI APIs alongside Codex: configurable base URL, optional Bearer API key, connection name and model/router combo ID.
  • One-click local 9Router preset, model discovery via GET /models, and manual IDs for APIs that omit discovery.
  • OpenAI-compatible Chat Completions and tool-calling models use the same precise, revision-checked, atomic and undoable editing tools as Codex. No shell or arbitrary file access.
  • Optional real image inputs and provider-compatible Low/Medium/High reasoning controls. Capabilities are explicitly enabled by the user, not guessed from model names.
  • Windows-encrypted saved keys, session-only keys, endpoint-scoped credential reuse and Forget connection. Keys stay out of project backups and public model responses.
  • Serial tool execution, Stop, request timeouts, bounded conversation/tool loops, no automatic retries and actionable authentication/quota/model errors.
  • AI editor topbar entry, clear provider/permission/cost information, and a native adapter loaded only for custom APIs.

Setup: open AI editor → Custom API, use the 9Router preset or enter a compatible endpoint, add the API key if required, then discover models or enter its exact model/combo ID. Start/configure your router separately. Enable images/reasoning only for supported models. Codex account sign-in remains available.

Cutline remains free and watermark-free. Custom providers may charge and receive the prompt/project context and permitted frames/transcripts; raw audio stays local. Compatibility requires OpenAI-style Chat Completions plus function/tool calling, not every proprietary API. Manual IDs are verified on the first message. Local mock integration tests do not prove every upstream model/provider works.

Download Setup to install or Portable to run without installing. Windows 10/11 x64 builds are unsigned and may trigger SmartScreen; verify SHA256SUMS.txt. This release does not automatically install or change your existing projects.

Verified: 97 unit/component checks, 31 native/protocol checks (including 10 custom-API tests), and 96 rendering/playback/export checks. The actual Windows window/preload/API/tool bridge passed a ten-point local mock workflow: discovery, encrypted storage, key privacy/clearing, title edit, real preview image, Undo, Forget connection and retained Codex setup. No real provider credentials or paid inference calls were used.

Cutline v0.8.6 — Timeline/audio synchronization

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 20:46

Cutline 0.8.6 — Timeline/audio synchronization

  • Rapid scrubbing keeps the latest seek target while media is still decoding.
  • Play waits for the final source position, audio initialization, a prepared preview frame and media playback readiness. Obsolete startup work cannot resume an old seek.
  • The playhead follows actual media playback, preferring audible audio, with trims, clip placement and speed accounted for. It no longer runs on an independent wall timer while audio is loading.
  • Decoder seeks and buffering pause the whole transport, keeping overlapping songs and video together. Drift correction uses 40 ms of timeline time instead of the former 180 ms source-time allowance.
  • Silent-gap playback stops at the next source boundary rather than skipping short clips after a late UI frame.
  • Playhead timestamps are read after preview rendering and committed immediately at a bounded cadence to avoid delayed pre-stall UI updates.

Validation: 97 unit/component checks, 21 native/protocol checks and 96 render/playback/export checks passed, plus the server-rendered shell (215 total). The final six transport checks were rerun after the UI timestamp change. Native Windows drag/release → Play and deliberate 100 ms UI-stall checks passed twice consecutively with generated media in an isolated profile. Desktop/web builds, type checking and scoped lint passed. These tests do not measure physical speaker or Bluetooth latency.

This addresses persistent internal synchronization drift, not all preview frame drops, hardware latency or unsupported codecs. An overloaded renderer cannot draw while blocked; it catches up to the media clock on its next update. Offline frame-accurate and disk-backed Windows export remain unchanged.

Download Setup for installation or Portable to run without installing. Windows 10/11 x64 builds are unsigned and may trigger SmartScreen. Verify downloads using SHA256SUMS.txt. Installing an update preserves local projects and original media; no automatic installation is performed by this release workflow.

Cutline v0.8.5 — Disk-backed Windows export

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 11:10

Cutline 0.8.5 — Disk-backed export

The biggest practical reliability gap identified in Cutline was output-size memory growth: the entire encoded video lived in RAM and was copied again through desktop IPC. This release adds automatic disk-backed export on Windows.

  • Pick the destination before rendering. Cancelling the save dialog starts no export.
  • MP4/H.264 and WebM/VP9/VP8 output writes in chunks of at most 1 MiB. Acknowledged disk writes apply backpressure to encoding; the finished file is not rebuilt in renderer memory or copied wholesale through IPC.
  • An exclusive temporary file is created beside the chosen destination. Native code validates session ownership, positions, chunk sizes and final file size. Encoding/frame validation and disk flush finish before an atomic same-directory rename replaces the approved destination.
  • Cancel and ordinary errors preserve existing videos and remove temporary output. Orderly shutdown waits for cleanup or an already-started final commit. Full/locked drives give actionable errors. The success view shows the saved location and size.
  • No extra runtime/model download, paid service or export watermark. Resolution, frame rate, effects and audio mixing are unchanged.

Verification: 93 unit/component checks, 15 font/Codex protocol checks, 6 native disk-writer checks and 90 render/export checks pass, plus the server-rendered shell (205 total). The production Windows UI, including the packaged app archive, streams exactly 960 frames at 60 fps with audio, starts disk writes before rendering finishes and limits chunks to 1 MiB. Destination cancellation, mid-export cancellation and simulated writer failure preserve the existing destination. Actual streamed files are decoded in all three formats under slow-writer backpressure. Tests cover seek-back header writes, cross-session rejection, final-size failure, shutdown races, full-drive errors and offsets beyond 4 GiB. The large-offset check is not a multi-gigabyte encoded video or a memory benchmark. Desktop/web builds, type checking and scoped lint pass; earlier live AI and Whisper checks were not repeated.

Limitations: browser downloads still assemble output in RAM; use the Windows app for disk-backed output. Decoded sources, codecs, surfaces and container metadata still need memory. MP4 is regular/seekable with metadata at the end, not Fast Start optimized for progressive web playback. Large projects still require adequate disk space and processing time; this does not provide proxy preview or repair choppy source footage. Forced termination/power loss may leave a .cutline-export-*.partial beside the destination; it is not published as a finished video. Final atomic commit cannot be cancelled. If temporary cleanup fails, the UI names the folder.

Windows x64 builds are unsigned and may trigger SmartScreen. Verify Setup/Portable downloads with SHA256SUMS.txt. Projects and original media are not changed by installing an update.

Cutline v0.8.4 — Frame-accurate export and library cleanup

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 10:26

Cutline 0.8.4 — Frame-accurate export and library cleanup

Cutline stays free, local and watermark-free. This release replaces live-recording export with offline frame rendering and adds safe media/project deletion.

  • Constant-frame-rate export: each frame renders at an explicit timeline timestamp. Slow effects or decoding take longer instead of silently dropping frames. Encoding uses backpressure and validates packet count and timing before saving.
  • Source-accurate rendering: independently decoded video frames support trims, speed changes, freezes, repeated sources and two-sided transitions. MP4/H.264 and WebM/VP9/VP8 are supported where codecs are available.
  • Synchronized audio: bounded-window mixing preserves volume automation, fades, stereo pan, muted layers and transition crossfades. Basic pitch-preserving speed processing is included.
  • Delete media: trash buttons in Media and Audio remove assets and linked clips in one undoable edit, with confirmation and locked-layer protection. Original files and other projects remain untouched.
  • Delete projects: Project → My projects lets you permanently remove a saved project, with confirmation. Shared assets remain; only unused imported copies are reclaimed. Deleting the open project creates a fresh workspace, and stale autosaves cannot recreate deleted projects.
  • Cancel export: cancellation disposes decoding/encoding resources without saving a partial result. Unsupported codecs fail explicitly instead of producing missing frames or silently absent audio.

Verification includes 93 unit/component checks, 15 font/Codex protocol checks and the full 89-check render/export suite. Export stress tests cover all three formats at 30/60 fps under deliberate rendering stalls, including decoded-audio beat synchronization. A production Windows UI export from the packaged app saved exactly 960 frames at 60 fps over 16 seconds with audio; native saving and cancellation passed. Media/project deletion is tested through real IndexedDB and the packaged production UI in isolated profiles, including Undo, project reset and layout at 1100/1480 pixels. Type checking, scoped lint, desktop/web builds and server rendering passed. Earlier live AI and Whisper model checks were not repeated.

Limitations: this prevents render-load-induced dropped frames, not every cause of choppy playback. It does not repair already-stuttering source footage or interpolate low-frame-rate footage. Demanding files still need a capable playback device. Export files are currently assembled in RAM; long/high-bitrate projects can use substantial memory. Extreme audio retiming can introduce artifacts. Keep the app open and PC awake while exporting. Project deletion is permanent: back up first.

Download Setup to install or Portable to run without installation. Windows x64 builds are unsigned and may trigger SmartScreen. Verify downloads with SHA256SUMS.txt. Optional Codex editing uses your own account and usage limits.

Cutline v0.8.3 — Text animations that move your way

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 08:06

Cutline 0.8.3 adds eight new text animation presets and much deeper animation customization. Core editing/export remain free, local and watermark-free.

  • New text presets: Letter Fade, Letter Slide, Letter Blur, Letter Spin, Letter Bounce, Letter Flip, Word Pop and Line Slide, on both entrance and exit.
  • Animate letters, words or lines: forward, reverse, center-out, edges-in or reproducible shuffled order; adjustable stagger and seed.
  • More motion control: angle/distance, pop overshoot, flip axis, zoom direction/strength, spin, blur, fade and Back/Spring easing where applicable.
  • Independent stack timing: each entrance/exit layer has a start delay and active span within its phase. Stack sequences with whole-clip motion, Combo loops and effects.
  • Better workflow: animation search, a scrollable preset grid and a reset that changes only the selected animation. Save/reapply the complete customized recipe.
  • Stable typography: reveals retain full-title alignment and preserve emoji/combining accents. Long titles batch adjacent units to bound work.
  • Persistence and AI editing: the new settings survive custom presets, project backups and splitting and are exposed to the existing validated Codex tools. Text-only presets are not offered on video.

Verification: all 103 unit/component/protocol checks and 81 rendering/export checks pass, plus the server-rendered shell (185 total), type checking and scoped lint. Checks include real preset/customization pixels, mixed stacks, native controls and saved recipes, and actual encoded entrance/exit frames. Production Windows UI, preview and small-window layout pass against the packaged app archive in an isolated profile. Earlier live AI and Whisper model checks were not repeated.

Known limits: real-time export can drop frames under heavy load; very short silent static-image edits can fail. Long/4K projects and large animation/effect stacks need device-specific testing. Automatic scene and beat detection remain estimates; object detection is bounding-box based, not subject tracking.

Download the installer or portable EXE below. Windows builds are unsigned; SmartScreen may show a warning. Verify downloads with SHA256SUMS.txt. Optional Codex editing uses your account and usage limits.

Website: https://cutlinevideo.vercel.app/
Interface screenshots and detailed documentation: https://github.com/ahpah-dev/Cutline---Open-source-video-editor/blob/v0.8.3/README.md

Cutline v0.8.2 — Scene cuts & a redesigned website

Choose a tag to compare

@ahpah-dev ahpah-dev released this 08 Oct 07:13

Cutline 0.8.2 adds local scene-cut detection and a redesigned introductory website. It includes the existing free, local-first editor, Codex media tools, automatic beat detection and workspace themes.

  • Reasoning effort: Auto and the reasoning levels actually advertised by the chosen model/account. Each turn receives the selected effort, including reused chats. Unsupported choices fall back to Auto when changing models; higher effort may increase latency/account usage.
  • Reference images: attach up to three local visual references in the composer. Images stay outside the timeline, are downsampled to at most 1024 px with original aspect ratio, and are bounded in size. Source images and original video frames are available through a dedicated inspection tool, separate from the edited preview.
  • Free local rhythm analysis: Codex can request 0.25–120 second windows of an imported audio/video clip. Local workers return estimated tempo/confidence, beat/onset times, real levels, clipping and silence, with source/timeline timing that respects trim and playback speed. Analysis does not change the project or automatically insert markers/captions.
  • Optional local speech: enable Whisper under Media access to authorize a one-time download on the first speech request. Tiny multilingual/English works on CPU (~50–100 MB); Large v3 (~1 GB) uses the existing WebGPU path and requires a compatible GPU. The assistant receives transcript/timing metadata, not raw audio. No paid audio provider or new API key is added.
  • Explicit limits: this is speech/rhythm analysis, not native audio hearing. Models on this Codex connection accept text/images, not raw audio. Recognition/beat estimates can be wrong and do not establish instruments, music mood, sound-effect identity or speakers.
  • Media controls and safe cancellation: disable media requests at any time; speech is off until enabled. Stop/disconnect/project changes terminate local analysis, and stale results are rejected. The composer remains accessible in small Windows layouts.
  • Earlier 0.8.1 media verification: actual account connection and supported-level discovery without inference, production Windows source-image/video-frame and reference-input tests, measured rhythm analysis, real Tiny speech transcription on synthetic speech, permission refusal/revocation and cancellation. Large v3 GPU inference was not separately tested; no claim of flawless editing is made.
  • Optional Codex editing still uses your account and usage limits. Requested source/preview images, reference images, project metadata and permitted transcripts/audio statistics are sent through Codex. Raw audio/original files stay local. Core editing and export remain free and watermark-free.
  • Known limits: real-time export; some heavy raster effects use a 1024-pixel processing surface. Long/4K projects require device-specific testing. Precise automatic cutout/tracking, freehand masks, optical flow, proxies, multicam and offline rendering are not implemented.
  • Known encoder limitation: very short silent static-image edits (under one second) can fail. Extend those edits to at least two seconds; the error includes this guidance.

New in 0.8.2: local scene-cut detection, inspired by CapCut Split scene and Premiere Scene Edit Detection. Scan a video clip, adjust sensitivity/minimum shot length, review before/after thumbnails, then split into editable scenes or create moment markers. Application is one undoable batch; trim/speed and keyframe/source-audio timing are preserved. Locked/frozen clips and stale results are rejected. No model, API key or media upload is required. The detector targets hard cuts; gradual dissolves and very brief shots may be missed. Scans are limited to ten minutes and 300 cuts. Review the estimates before applying.

The introductory website has been redesigned with seven actual interface views, full-size screenshot dialogs, responsive feature cards, accessible keyboard navigation, local fonts, reduced-motion support and current installer/portable/checksum links. Live: https://cutlinevideo.vercel.app/

0.8.2 verification: 96 unit/component/protocol checks, 76 render/export checks and the server-rendered shell (173 total), type checking and scoped lint pass. Packaged production scene detection/split/undo/cancel/layout checks pass. Website gallery/keyboard/dialog/FAQ checks pass at desktop, mobile and 200% zoom without horizontal overflow or browser console errors. Earlier model/media checks were not all repeated.

Download the installer or portable EXE below. Both are free and watermark-free. These Windows builds are unsigned, so SmartScreen may show a warning. Check SHA256SUMS.txt to verify downloads.

Interface screenshots and detailed documentation: https://github.com/ahpah-dev/Cutline---Open-source-video-editor/blob/v0.8.2/README.md

Cutline v0.8.1 — Codex reasoning, images and local audio analysis

Choose a tag to compare

@ahpah-dev ahpah-dev released this 07 Oct 17:19

A more capable Codex video assistant for the free, local-first Windows video editor. Includes all existing editing, automatic beat detection and theme features.

  • Reasoning effort: Auto and the reasoning levels actually advertised by the chosen model/account. Each turn receives the selected effort, including reused chats. Unsupported choices fall back to Auto when changing models; higher effort may increase latency/account usage.
  • Reference images: attach up to three local visual references in the composer. Images stay outside the timeline, are downsampled to at most 1024 px with original aspect ratio, and are bounded in size. Source images and original video frames are available through a dedicated inspection tool, separate from the edited preview.
  • Free local rhythm analysis: Codex can request 0.25–120 second windows of an imported audio/video clip. Local workers return estimated tempo/confidence, beat/onset times, real levels, clipping and silence, with source/timeline timing that respects trim and playback speed. Analysis does not change the project or automatically insert markers/captions.
  • Optional local speech: enable Whisper under Media access to authorize a one-time download on the first speech request. Tiny multilingual/English works on CPU (~50–100 MB); Large v3 (~1 GB) uses the existing WebGPU path and requires a compatible GPU. The assistant receives transcript/timing metadata, not raw audio. No paid audio provider or new API key is added.
  • Explicit limits: this is speech/rhythm analysis, not native audio hearing. Models on this Codex connection accept text/images, not raw audio. Recognition/beat estimates can be wrong and do not establish instruments, music mood, sound-effect identity or speakers.
  • Media controls and safe cancellation: disable media requests at any time; speech is off until enabled. Stop/disconnect/project changes terminate local analysis, and stale results are rejected. The composer remains accessible in small Windows layouts.
  • Verification: unit/protocol/render/export regression checks, actual account connection and supported-level discovery without inference, production Windows source-image/video-frame and reference-input tests, measured rhythm analysis, real Tiny speech transcription on synthetic speech, permission refusal/revocation and cancellation. Production media checks are repeated against the packaged archive. Large v3 GPU inference is not separately tested in this update; no claim of flawless editing is made.
  • Optional Codex editing still uses your account and usage limits. Requested source/preview images, reference images, project metadata and permitted transcripts/audio statistics are sent through Codex. Raw audio/original files stay local. Core editing and export remain free and watermark-free.
  • Known limits: real-time export; some heavy raster effects use a 1024-pixel processing surface. Long/4K projects require device-specific testing. Precise automatic cutout/tracking, freehand masks, optical flow, proxies, multicam and offline rendering are not implemented.
  • Known encoder limitation: very short silent static-image edits (under one second) can fail. Extend those edits to at least two seconds; the error includes this guidance.

Download the installer or portable EXE below. Both are free and watermark-free. These Windows builds are unsigned, so SmartScreen may show a warning. Check SHA256SUMS.txt to verify downloads.

Interface screenshots and detailed documentation: https://github.com/ahpah-dev/Cutline---Open-source-video-editor/blob/v0.8.1/README.md

Cutline v0.8.0 — Automatic beats and workspace themes

Choose a tag to compare

@ahpah-dev ahpah-dev released this 07 Oct 16:11

Advanced local beat marking and four complete workspace themes for the free, local-first Windows video editor. Also includes the previously local-only 0.7.1 text-font fix.

  • Auto beats: open the toolbar button beside Mark beat, choose an audio/video clip and Detect beats. Local spectral-flux onset detection, tempo estimation and tempo-constrained tracking run in a cancellable worker. No model download, subscription or footage upload.
  • Refine before applying: waveform/marker preview, rhythm beats or strong-hit detection, sensitivity, BPM override, every/other/fourth/half-beat density and millisecond timing offset. Half-beats are interpolated subdivisions. Source trim and playback speed are respected; analysis covers the first 10 minutes of the selected edit.
  • Safe marker batches: frame-aligned, deduplicated, one-step undo. Manual beats/moments and other sources remain intact. Reanalysis can replace only the selected source clip's previous automatic guides. Backups and Codex snapshots retain source-clip/BPM metadata. Markers are absolute guides; reanalyze after moving/trimming/changing speed.
  • Full workspace themes: Graphite, Midnight, Forest and Light include the custom Windows topbar, panels, timeline, menus and controls. The device preference survives restart and is independent of projects. Footage and exported colors are unchanged.
  • Text-font mismatch fixed: selected fonts load before preview/export, paused frames redraw, and non-Latin subsets and font-style keyframes are prepared. The Manrope inspector label now matches the actual rendered font.
  • Beat detection is music-dependent: confidence is an estimate, and sparse recordings or half/double-time rhythms can need adjustment. This does not auto-cut video, classify downbeats or guarantee perfect timing.
  • Verification includes 155 automated model/component/protocol/render/export checks, production Windows beat analysis/undo/cancellation and theme persistence/layout checks, repeated against the packaged app archive. These checks are not a guarantee of flawless editing on every device.
  • Optional Codex editing uses your account and its usage limits. Project metadata and requested preview frames are sent to Codex; original media stays local. Core editing and export remain free and watermark-free.
  • Known limits: real-time export; some heavy raster effects use a 1024-pixel processing surface. Long/4K projects require device-specific testing. Precise automatic cutout/tracking, freehand masks, optical flow, proxies, multicam and offline rendering are not implemented.
  • Known encoder limitation: very short silent static-image edits (under one second) can fail. Extend those edits to at least two seconds; the error includes this guidance.

Download the installer or portable EXE below. Both are free and watermark-free. These Windows builds are unsigned, so SmartScreen may show a warning. Check SHA256SUMS.txt to verify downloads.

Interface screenshots and detailed documentation: https://github.com/ahpah-dev/Cutline---Open-source-video-editor/blob/v0.8.0/README.md

Cutline v0.7.0 — Studio redesign, color grading and local object detection

Choose a tag to compare

@ahpah-dev ahpah-dev released this 07 Oct 10:06

A redesigned free, local-first Windows video editor. This release includes the previously unreleased 0.6.0 library expansion as well as the new 0.7.0 grading and detection tools.

  • Studio redesign: graphite/slate surfaces, individually framed panels, larger labels, a wider inspector, readable two-row property navigation, compact preset cards and a clearer color-coded timeline. Offline Inter typography, keyboard focus and reduced-motion support.
  • 34 effects and 44 transitions: categorized search, actual renderer thumbnails, live hover/keyboard-focus previews and ordered effect-stack controls. Both clips participate in transitions, with exact endpoints and consistent preview/export.
  • Advanced color grading: exposure, tint, tonal range, vibrance and hue, three color wheels, master/R/G/B curves, numeric point editing, viewer luma/RGB histograms and non-destructive bypass/reset. Scalar properties support keyframes and validated AI editing.
  • Grading uses a full-resolution GPU pass with an alpha-preserving CPU fallback. It is display-referred 8-bit sRGB/Rec.709, not HDR/RAW. Curves are clip-wide, not keyframed.
  • Local object detection: explicit consent before downloading roughly 10 MB of free YOLOS Tiny weights. Find common objects in the selected source frame, then create an editable rectangle mask or source crop. No footage upload, account or API key. Cached model inference works offline. This is bounding-box detection, not precise subject cutout or tracking.
  • Chroma-key enabling fixed: clicking the visual switch or label changes the checkbox rather than accidentally triggering its keyframe diamond. Saved state, actual keyed pixels, undo and layer locking are regression-tested.
  • Retains editable masks, seventeen blend modes, cropping, center snapping, beat/moment markers, layer locking, detach audio and keyframeable stereo pan.
  • Verification includes 148 automated model/protocol/render/export checks plus production Windows layout checks at 1100×700, 1480×920 and 1920×1080, real detector inference and fresh-worker offline cache testing. These checks are not a guarantee of flawless editing on every device.
  • Optional Codex editing uses your account and its usage limits. Project metadata and requested preview frames are sent to Codex; original media stays local. Core editing and export remain free and watermark-free.
  • Known limits: real-time export; some heavy raster effects use a 1024-pixel processing surface. Long/4K projects require device-specific testing. Precise automatic cutout/tracking, freehand masks, optical flow, proxies, multicam and offline rendering are not implemented.
  • Known encoder limitation: very short silent static-image edits (under one second) can fail. Extend those edits to at least two seconds; the error includes this guidance.

Download the installer or portable EXE below. Both are free and watermark-free. These Windows builds are unsigned, so SmartScreen may show a warning. Check SHA256SUMS.txt to verify downloads.

Interface screenshots and detailed documentation: https://github.com/ahpah-dev/Cutline---Open-source-video-editor/blob/v0.7.0/README.md