Release notes

What's new in KnoxStudio

Follow every release as the app evolves — from timeline editing and AI generation to export, screenwriting, and performance work.

Development statusActively maintained · Latest v1.3.2

Implemented audio-only export (.mp3 and .wav) end-to-end
Update carmera bubble lives
Update multi-segment paths:
- the concat path (every clip clamped to [cv{i}] before concat),
- the xfade/transition path (each [sv{i}] clamped, which also makes the transition offsets — derived from seg.duration — line up with the real stream length).

Separate system audio and microphone audio on the timeline tracks
Remove tokens usage
Auto-naming chat history title better
Add "Back to chat" link
Use stable_slider for all Sliders to replace default egui ones
Redesign slider visuals (rounded track, teal progress fill, circular handle with hover/drag emphasis) and unify every inspector slider (Volume, Fade In/Out, Speed, Opacity, Scale, Rotation) onto one framed component with an inline label + editable value, removing duplicate value displays

Editor mode — AI timeline editor

Editor agent is now a full timeline editor that can do everything a human can on the tracks: split, trim, move, delete, retime, opacity/volume/fade/transform, add/remove transitions, duplicate, copy/cut/paste, detach audio, freeze frames, lock/unlock, group/ungroup, ripple-delete a range, and insert space.
Move the red playhead/scrubber, select clips, and manage tracks (add, remove, rename, mute/solo/lock/volume) directly from chat.
Frame extraction from real footage: extract a frame (playhead/first/last/specific time) or auto-detect key frames and drop them on the timeline as stills.
Insert media from a file path, an `@`-mentioned attachment, or the Media Library.
AI generation from the timeline: `regenerate_clip`, `generate_variation` (opens the variations panel), and `extend_clip` queue async generations grounded in the source clip.
Vision-grounded edits: the agent can watch the actual footage (by clip, time, or playhead) before deciding where to cut/trim.
Selection-aware: the agent now sees what you have selected, so "slow this down", "delete this", or "caption the selected clip" resolves to the right clips.
Richer timeline awareness: the snapshot now shows canvas resolution, per-clip lock/group/audio/transition/transform state, per-track gaps, and markers — so edits like removing a transition or closing a gap are grounded, not guessed.
Content analysis without watching frames: `analyze_timeline` returns scene-change cut points and silent/dead-air ranges so the agent can "remove the silences", "tighten the pauses", or "cut on every scene change" precisely.
Edit by what's said: `transcribe_clip` transcribes a clip's speech (ElevenLabs Scribe v2) with timeline-time timestamps, so the agent can "cut to where she says X", "remove the part about Y", or caption the dialogue.
Titles & captions: `add_text` places a text overlay (title card, caption, lower-third) on the canvas; `add_marker` / `remove_marker` manage chapter/beat markers.
Safety: destructive edits (delete / ripple-delete / remove track) prompt read-back verification, and in Confirm mode they wait for an in-chat approval card before being applied. If you edit the timeline while the agent is working, its changes are held for your approval instead of overwriting your manual edits.
Director mode support vision model to watch the previous clip, then extract a continuity frame for next video generating base on contexts and prompts

Add custom modal
Return the directory a file-import dialog of the given `key` should open in
Adjust unsave project open new one modal optimization
Improve click effect playing dynamically even speed changes
Match exported click effect to the preview: animate the ring (grow + fade) as a per-frame sequence, fix overlay alignment in original-resolution exports, and render the preview ring in project (compose) pixels so its size matches the exported video
Export filename now derives from the project name instead of the hardcoded export
Redisign the recording setup modal

LLMs supporting for Screenplay editor
Git supporting for Screenplay editor
Remove unused accessibilities
Director mode: AI director shot breakdown with framing, camera, and per-shot timing
Director mode: Interactive Shot List editor to review and edit prompts and durations before generation
Director mode: Send screenplay directly from the Screenplay editor toolbar
Director mode: Syllable-accurate duration enforcement with auto-split for overlong dialogue
Director mode: Robust planner JSON parsing with one repair round before fallback
Director mode: Persist final shot plan snapshot into chat history when the pipeline starts
Director mode: Accurate duration totals when skipping failed shots
Add fallback：Hiragino Sans GB → STHeiti Light
Chat history items list optimizations
Fix some bugs

App Optimization

Add Media Preview
Add Media Library Sidebar TreeView
Improve the Timeline Zoon actions
Update all phrases for multilingual support with Chinese & English
Update Inspector Icons
Enhance Screenplay Editor

Screenplay Editor — seamless workspace integration

The screenplay editor is no longer a floating window. It now renders inside the central
The editor's separate file sidebar was merged into the Media Library: clicking an editable
JSON and YAML files open in the same inline editor with data-aware syntax highlighting in Edit
Media Library context menus gained screenplay actions (Open, Rename, Duplicate, Pin/Unpin,
Focus mode (Cmd+Enter / F11) hides the Media Library and right panel for distraction-free
The last-open screenplay document is restored on launch.
Read-mode scene cards now show a "Roles:" chip row mapping each scene character to its
Scene cards with generated media gained an "↗ Add to timeline" action that imports the scene's
Screenplay file operations (new / rename / delete / duplicate) now refresh the Media Library

Removed

Pruned the screenplay editor's retired internal file tree (`file_browser.rs` and the unused

Added

Video Continuity System

Video sequence and shot tracking for multi-clip continuity across scenes
Automatic frame chaining — exit frames extracted and injected as entry frames for the next shot
Scene environment propagation and prompt enhancement across all shots in a sequence
Motion state analysis via prompt heuristics and Gemini 3.5 Flash vision (KnoxChat or Google API)
ContinuityContextBuilder unifying frames, environment, and motion for generation
SQLite persistence for sequences with auto-migration from JSON and real-time state saves
Sequential shot generation integrated with JobQueue, Director, and batch scene generation
ContinuityAwarePlan wrapper for multi-step video plans with automatic frame chaining
Failed shot tracking with per-shot and batch retry support

Agent Panel & Sequence UI

Sequence list view (chain link icon) with pause, resume, delete, and duplicate controls
SequenceCard with real-time progress, ETA, and generation statistics
Shot preview popovers with exit-frame thumbnails, status, and error details
Timeline mini-map with duration-proportional, color-coded shot blocks
Inline shot editing — prompt, duration, resolution, aspect ratio, and audio toggle
Shot reordering, add-shot to existing sequences, and batch delete failed shots

Timeline Integration

Send to Timeline — one-click import of completed sequences to a new video track
Import preview dialog showing shot timings and total duration before committing
Continuity chain markers and connecting lines between sequence clips on the timeline
Quick jump from a completed shot to its clip on the timeline

Settings & Notifications

Continuity toggles in AI Agent Config → Defaults (frame chaining, sequential mode, vision analysis, environment sync, motion prompts)
macOS desktop notifications when sequences complete or fail, with preference toggle
Escape shortcut to close sequence view popovers

New generated content (videos, images, audio) will now be added after the last clip on the timeline
Multiple generated assets are placed sequentially (each after the previous one)
No overlaps - content is automatically appended to the end
Explicit positions from the timeline_place tool are still respected when specified
User-initiated placements (like variations panel) continue to use playhead position as intended

Added

Architecture documentation (`docs/ARCHITECTURE.md`)
Developer onboarding guide (`docs/DEVELOPER.md`)
Architecture Decision Records (ADRs) in `docs/adr/`
Keyboard shortcuts reference (`docs/KEYBOARD_SHORTCUTS.md`)
FFmpeg license compliance documentation

Added

AI & Generation

Manager Agent with streaming responses and tool calling
Voice agent support via GPT-4o Realtime API
Multi-provider video generation (Minimax, Luma, Kling, Veo)
Character role system with reference images
Screenplay editor with scene-driven generation
Generation history with variations support
Cost tracking dashboard for AI usage

Data Integrity (Phase 7)

Project file backup rotation (keeps last 5 autosave snapshots)
SHA-256 checksum for project bundle corruption detection
SQLite database backup on app launch
Export output verification with duration/stream checks
Preferences migration system with versioned schema
Configuration schema validation with helpful error messages
MDM/enterprise configuration profile support

Internationalization (Phase 6)

Complete English and Chinese translations (650+ keys)
Translation completeness validator
Pluralization support (`t_plural!` macro)
Locale-aware date/time/number formatting
High-contrast theme variants (dark and light)
VoiceOver accessibility labels
Full keyboard navigation support

Observability (Phase 4)

Crash reporting with local storage
Export performance metrics (FPS, ETA)
AI generation cost tracking
Health check system (ffmpeg, disk, memory)
Session analytics (opt-in)
Runtime log level toggle
Export progress ETA calculation

Build & Release (Phase 5)

Comprehensive Makefile with all targets
Universal Binary build automation (ARM64 + x86_64)
Automated notarization workflow
Auto-update system with GitHub releases
Version bumping script
SBOM generation
Pretty DMG creation with custom layout
Changelog generation from commits

Performance (Phase 3)

Incremental undo with diff snapshots
Memory pressure monitoring with adaptive cache eviction
Frame buffer pool for video player
Performance benchmarks (Criterion)
Lazy track loading for large projects
Batch annotation rendering for export
Async thumbnail generation pipeline
Zstd compression for project files

Testing (Phase 2)

Export pipeline integration tests
AI agent tool dispatch tests
Fuzz testing for project file deserialization
Property-based tests for timeline operations
Rate limiter edge case tests
Snapshot tests for FFmpeg filter graphs
Circuit breaker for external APIs
Undo/redo stress tests
Crash recovery tests
Code coverage reporting (cargo-llvm-cov)

Foundation (Phase 1)

Pre-release check script (build, test, clippy, fmt)
`cargo audit` integration
macOS Keychain for API key storage
Retry logic with exponential backoff
Log rotation (10MB max, 5 files)
Safety documentation for all `unsafe` blocks
Input sanitization for LLM prompts
Graceful shutdown with WAL checkpoint

Changed

Improved timeline rendering performance
Better error messages for common issues
Modernized preferences dialog

Fixed

Memory leak in video player frame copy
Timeline zoom persistence across sessions
Audio waveform rendering at high zoom levels

Added

Core Features

Screen recording with ScreenCaptureKit
Multi-track timeline editing
Video/audio clip management
Annotation tools (shapes, text, arrows)
FFmpeg-based export pipeline
Project file format (.knoxstudio bundles)

UI

egui-based interface
Canvas preview with annotations
Timeline with tracks, clips, markers
Inspector panel for clip properties
Dark and light themes

Recording

Screen capture with cursor
Microphone audio recording
System audio capture
Camera overlay recording
Click highlight effects

Export

MP4 (H.264) export
WebM (VP9) export
GIF export
Custom resolution and bitrate
Audio mixing

Project Management

Autosave snapshots
Recent files list
Undo/redo history
Project templates

Technical

Rust + Swift FFI architecture
AVFoundation video playback
SQLite for media indexing
Tokio async runtime for AI

Release timeline

Major milestones at a glance.

Version	Date	Highlights
v1.1.0	2026-05-23	AI agents, data integrity, i18n
v1.0.0	2026-03-15	Initial release