1.1.25

New

• Tracking: a motion layer can follow a track
• Tracking: follow faces and objects through footage, with MCP tools
• Chat: select, text, any colour and line width in the frame markup editor
• Chat: @time references and frame attachments with notes and markup
• Thumbnail: edit the whole prompt before the thumbnail project is made
• Export: a time left that follows the recent pace, a range when it is a guess, and the export's settings in the progress dialog
• Projects: choose to keep or delete a project's folder when removing it
• Editor: Generate thumbnail button creates a thumbnail image project from the video
• Thumbnail: create a thumbnail image project from a video project, with fakes-driven tests
• Thumbnail: pure frame choice, headline and brief logic for video-to-thumbnail projects
• Export: project resolution, 4K/2K/1080p/720p export sizes from it, and H.265 file type
• Export: resolution choices, render steps with a time estimate, and a Cancel that acts at once
• Export: draw motion layers in the background so a full export finds them ready
• Import: add several files, whole folders and drops to a project at once
• Engine: hold the Gemini loop to a measured review after it edits
• Engine: Gemini runs an agent loop over FrameFlow's MCP tools
• Chat: New conversation button for the internal CLI agent
• Engine: the internal Claude agent continues its conversation
• Engine: a CLI always runs a named model and effort
• Engine: list the newest Claude models by version; label the default plainly
• Agent: drop editing personas, one shared baseline, per-engine model + effort
• MCP: inspect_audio lets agents see any sound

Fixed

• Projects: the one-time import takes the newer copy of a project both folders have
• Settings: a launch that cannot read a stored key no longer erases it
• Tracking: a following cover sits on the frame the export shows
• Editor: bars hold together in a narrow window; the window has a minimum size
• Chat: the composer row keeps Send on screen; setup describes Gemini as it works
• Projects: every unpackaged launch uses one data folder, so one project list
• Projects: a project removed with its files kept stays removed
• Chat: the Text tool opens a field on a real click; Draw is the default tool
• Chat: copy selected text, readable selection in your own bubble, copy button
• Export: one colour space for the whole file (SDR BT.709) and a tag that is true
• Editor: selecting a clip leaves the selected transition
• Editor: narrow media panel rows, capped clip tray, audio/mpeg for Gemini transcripts
• Baseline: keep the inspect_audio sentence free of banned opinion words
• Chat: remove the pinned "Applied V#" result card

Faster

• Export: draw motion layers at 30 fps and overlap capture with the write
• Export: capture motion layers in parallel and at the proxy frame rate
• Export: fast preview at 480p, regions in parallel, faster caption pass

---

Installers are attached below. Full changelog: https://github.com/navidshad/frameflow/commits/v1.1.25

1.1.24

New

• Import: two lanes, words and picture, run side by side
• Proxy: hardware decode and GPU scaling for every source, two proxies at once
• Transcript: real progress, parallel word windows, Transcribe again button
• Editor: an Exports tab so finished exports can always be found
• Exports: human file sizes and dates for the exports list
• Exports: list a project's finished exports, reveal and open them
• Revisions: one revision per agent round
• Effects: switch each effect on and off to compare
• Multicam: speakers tests, honest unsure cutoff, real-footage scoring, docs
• Multicam: find who speaks when
• Multicam: sync cameras by sound and cut between angles
• MCP: a reconnected agent can resume its own round by id
• MCP: splitAt op; tidy the round footer in the chat
• Editor: export and revision switches wait while an MCP agent holds the project; render_frames keeps its frames
• MCP: inspect_media, render_frames, find_silences, sequenceFrame; install a missing yt-dlp
• MCP: agent rounds, one agent per project, MCP conversation in the chat
• Motion: project motion kits shared by custom pages
• Motion: bundle Vazirmatn for Persian and Arabic text
• Start: first-launch setup wizard and two-road Home
• Settings: rebuild Settings as a seven-section page
• Editor: chat box engine chip lists models
• Audio: provider slot — generated voice, effects and music
• MCP: sound tools and ops for agents
• Sound: synth palette, Sound tab, sound the motion
• UI: v2 graph — flat node panels, tool column, no ambient blobs
• UI: one dialog style — AppDialog for every modal
• UI: v2 settings — one column of two-column sections
• UI: v2 top bar on every page, in-app dialogs, Home and first run
• Audio: waveform, audio clip inspector, track roles
• Editor: v2 shell — flat surfaces, Inspect tab, full-height chat, preview rows
• Motion: agent-written pages are checked before they replace anything; plain ease() and clamp()
• Engine: Retry on a failed built-in engine run; drop Gemini CLI
• Motion: custom HTML authoring — code view, checks, last-passing, saved templates (board 5)
• Motion: determinism probe, saved templates, html ops with the MCP permission, AI context for custom pages
• Engine: ship Claude Code only; park Codex and Gemini CLI; render_frame burns text opt-in
• Motion: custom layers draw their last passing version; shared logic for checks, definitions and provenance
• Engine: engine chip, AI engine settings card, steps in the turn card, agent focus on the monitor
• Keyframes: keyframe lanes in the timeline (board 2)
• Engine: CLI engines run Claude Code, Codex and Gemini CLI through FrameFlow's MCP server
• Motion: bundle Outfit for template headings
• Keyframes: shared lane logic — lanes, curves, snapping, selection, mirror
• Motion: colour and enum settings can be keyed, and hold
• Motion: Motion library, M1 track, motion clips, monitor layer and inspector
• Motion: the editor AI can place and retime motion layers
• Motion: export motion layers through a hidden capture window
• Agent: Agent tab, editing banner, changed-by-agent outlines, Agent access card
• Motion: the motion layer model, page contract, templates and lint
• MCP: a local MCP server agents can read and edit the open project through
• Monitor: a title is edited on the frame like everything else
• Monitor: the frame tools say what they do
• Monitor: open on the edit, and put the transport by the timeline
• Timeline: drop past the edge to make a track
• Timeline: the lanes scroll while you drag
• Timeline: a drag says where it can land, and why not
• Timeline: video layers on top, audio below, and a way to add either
• Monitor: edit on the canvas — move, scale, rotate by the handles
• Preview: pin the exact frame to the playhead
• Effects: every control says whether the monitor can show it
• Timeline: say at edit time that a layer crosses a transition
• Preview: render the exact frame under the playhead
• Editor: an icon rail, and the toolbar says which key does what
• Editor: safe areas, a timecode you can type, and fullscreen
• Editor: a real keyboard, and a way to find it
• AI: reach layers from the ops schema
• Render: generated layers, and a way to delete a track
• Render: key a background out, so a layer sits in FRONT
• AI: suggest what to cut away to, from what is already known
• Editor: cut by what was said, not by where the clips are
• AI: let the editor reach everything phases 2 and 3 built
• Export: queue several formats instead of picking one
• Export: deliver ProRes as well as H.264
• Editor: keyframes — a parameter that changes over a clip
• Render: picture-in-picture, and a preview element to see it in
• Render: move, scale and rotate a clip in the frame
• Effects: packs, and the four looks that prove the loader
• Audio: duck a music bed under speech
• Editor: titles, through the pipeline captions already built
• Audio: repair the sound, and normalise the mix
• Editor: drag an effect onto a clip
• Render: resolve the effect seam into a real catalogue
• Dev: drive the running app, and fix what that found
• Editor: burn the transcript into the picture
• Editor: choose the frame, and reframe clips into it
• Render: let the sequence choose its own shape
• Editor: show transitions in the preview monitor
• Editor: transitions library and the junction drop
• Render: draw transitions, and place slices inside their region
• Render: plan transition overlap regions
• Electron 28 -> 44, and the packaging path to a signed installer
• drop the Python dependency and put CI behind the codebase

Fixed

• Chat: notes alone no longer bring the Applied card back
• Chat: drop the Applied card unless something needs the user
• Transcript: say when Gemini prepaid credits are depleted
• Agent: the editing bar follows the open round, not recent calls
• Preview: tag proxies so the monitor shows the export's pixels
• Preview: show Exposure as the addition ffmpeg makes
• Preview: show Warmth in the monitor, matching the export
• Multicam: an angle click changes the clip you chose, not its neighbours
• MCP: report interrupted media preparation instead of running forever
• Editor: a clip's filmstrip needs only the proxy
• Render: a cut near another track's edge no longer fails the export
• Multicam: switchAngles no longer folds punch-ins back into the wide shot
• Editor: ripple a clip that a cut trims from its very start
• Silence: read only the sound when finding pauses
• Multicam: label get_angles tiles with timeline time
• Editor: clear a wrongly failed sound even when it is not on the timeline
• Editor: no filmstrip for audio assets; it failed every synthesised sound
• Mix: sounds too short to meter no longer read as a quiet role; random round ids
• Motion: the probe accepts 1/255 raster noise; render_frames says the export-clips line once
• Motion: still frames draw a page with its kit
• Settings: save settings.json atomically
• Ops: removeRanges closes the spans it cuts, on every track
• Export: never hang silently on a stuck ffmpeg call
• Export: pad black with tpad instead of a color base
• Gemini: send generateText system instruction via config
• Export: pad each region's picture over an exact black base
• Transcript: measured word timing, current models, native script
• Render_frame: clearer wording while export clips render
• Export,render_frame: exact-length export joins and motion-aware exact frames
• Editor: store revisions and history by content hash; a big project always opens
• MCP: queue writes, summarise get_timeline, opt-in token file
• Motion: keep layer lengths, retime onto free tracks, explain probe failures
• Motion: motion-only timelines play, chat and export; pages follow their length; agents get a page guide
• Engine: an engine turn's failure shows once, in its chat card
• Engine: a CLI engine's HTML permission follows the user's setting
• Engine: render_frame shows titles and captions; CLI turns can revert; clearer errors
• Agent: badge a split cut by what was cut; close quiet sessions; clear the header
• Motion: lower third's accent bar no longer collapses on the 'bar' style
• Ffmpeg: repin BtbN builds to a month-end release that is not pruned
• Monitor: a generated layer moves with its transform
• Monitor: a generated layer stacks by its track, not under everything
• Monitor: the frame crops the picture again, handles or not
• UI: tooltips were being sliced off by the panel they sat in
• Inspector: sliders paint while you drag, commit once on release
• Editor: stop the monitor lying about gaps and doubled audio
• Editor: gain, fades, track gain and pitch are heard in preview
• Preview: key the transition cache on the graph, not a field list
• Render: a compositing layer's pad blacked out the track beneath it
• Editor: the shortcut list traps the keyboard while it is up
• Security: encrypt the API key at rest
• Editor: silence the outgoing clip once a transition ends
• Editor: stop the playhead looping at the end of a transition
• Editor: preview renders failed on every transition
• Editor: make the preview cache actually reach the monitor
• Editor: bind the transition slider, and free the trim handles
• Ffmpeg: read zip in-process, so the licence check runs on Linux
• replace ffmpeg-static with licence-verified FFmpeg 9.0.1 builds
• Ci: do not export an empty CSC_LINK on the macOS legs
• repair the lockfile so npm ci works, and delete the stale yarn.lock

Faster...

Release v1.1.23

Features

• Automatic updates for yt-dlp with a self-healing mechanism when stale-binary failures occur.

Bug Fixes

• Resolved issues related to stale yt-dlp binaries by automatically recovering and retrying.

Improvements

• More resilient update behavior, reducing manual intervention when yt-dlp binaries become outdated.

Release v1.1.22

Improvements

• Disabled the auto PR description CI workflow to prevent automatic PR description updates.

Release v1.1.21

Features

• Context sizing improvement (Editor): Size the context based on the source rather than intent, including a survey band plus a detailed pass.

Bug Fixes

• Editor build measurement fix: Ensure measureBuild properly handles op-less diffs to avoid answer turn shortfalls.

Release v1.1.20

Features

• None

Bug Fixes

• None

Improvements

• Updated documentation to lead with the timeline editor and use the original graph screenshot for the graph.
• Refreshed README screenshots with captures of the current UI.
• Corrected README documentation describing the retired video pipeline to reflect “pillar one.”

Release v1.1.19

Features

• Split editing outcomes: separate Home entry and keep video edits within the timeline.

Bug Fixes

• Fixed editor undo behavior: the main component now owns the undo ring cap (prevents duplication in the renderer).
• Corrected outdated documentation that referenced deleted code.
• Prevented item operations on an empty timeline and fixed related sizing/shortcut calculations.

Improvements

• Retired the AI video pipeline; the graph now serves images only.

Release v1.1.18

Features

• Enable long-form persona workflows to build full-length cuts.
• Add full-height toggle for right-side panels; timeline adapts accordingly.
• Allow drag-resizing of timeline height and panel widths.
• Support background exports from an in-memory timeline snapshot.
• Introduce handler-based render builders, including audio fades and improved export warnings.

Bug Fixes

• Fix ripple head-trim overlap, orphan clips, and description mis-pairing.
• Prevent re-transcription from destroying asset thumbnails.
• Stop storage location nesting a “FrameFlow” folder inside itself.
• Ensure configured model names are shown on token-spending buttons.
• Correct placement, coverage, and transcript “hole” defects found during audit.
• Make ripple trims/retimes behave correctly (ripple after retimes and trims, not only removals).
• Close timeline gaps both from AI actions and manual edits.
• When there are no projects yet, offer the video editor entry point.
• Follow route parameter changes on VideoEditorPage.
• Improve transcript reliability by chunking long audio and detecting incomplete transcripts.

Improvements

• Improve editor rendering/export behavior with handler-based builders and clearer export warnings.
• Enhance timeline usability with resizable layout controls and adaptive panel height.

Release v1.1.17

Improvements

• Updated auto-pr-description.yml configuration to improve automated pull request descriptions.

Release v1.1.16

Features

• Reworked the editor right rail into two independently-collapsible columns/rows (including Inspector + Chat split view).
• Added split right-rail layout with Inspector and Chat shown together.
• Added dense, batched filmstrip for timeline clips.
• Added timeline minimap and marker/chapter jump list for long-timeline navigation.
• Added scene detector sensitivity controls and improved merge/split piece corrections.
• Added per-asset transcript and silence/dead-air detection.
• Added auto-segmentation for audio assets (silence split with interval fallback).
• Enabled exporting/rendering via the timeline export/render engine.
• Expanded audio asset pipeline: import → preprocess → place → preview (and mix export for audio-kind assets).
• Added revision tree for AI edits (list + graph) and reworked proposal flow.
• Introduced AI prompt bar, editor personas, and diff proposals.
• Added timeline editing foundation with EDL preview and undo/redo.
• Added timeline video editor scaffolding and media pipeline.
• Added numeric progress indicators for every editor processing task.
• Added dense/batched navigation + UX updates for marker/chapter browsing.

Bug Fixes

• Matched right-column collapse controls to the media panel behavior.
• Prevented audio from reloading every frame; audio preview now free-runs without constant re-seeking.
• Improved audio playback performance by pooling audio elements and attaching them to the DOM.
• Played audio tracks in parallel using a per-track element pool.
• Fixed compact filmstrip/context toggle and improved vertical scrolling for tracks.
• Fixed prompt placeholders overflowing the narrow chat rail.
• Gated chat until media preprocessing is complete.
• Fixed marker jump-list popover rendering under the timeline playhead.
• Fixed timeline clip trim handles getting stuck after the first click.
• Improved prompt meta row layout to prevent label crowding.
• Reduced chat rail “chrome” for better text-level controls and one-frame-per-region behavior.

Improvements

• Drag-and-drop improvements: support dragging whole files onto the timeline and fixed audio-lane drops.
• Removed transcript/scene-description data from an asset (cleanup).
• Referenced local media in place instead of copying it.
• Speeded up proxy generation for large files.
• Improved editor project naming (auto from first import + inline rename).
• Added snappier popover animation (replaced slow slide-up).
• Added settings/UX gating for decorative animation behind prefers-reduced-motion.
• Updated UI dependency (pilotui 1.28.1 → 1.29.1).
• Added Playwright MCP artifacts to .gitignore and added Playwright MCP for driving the Electron UI.