SceneForge Studio 0.9.4-beta.9

SceneForge Studio 0.9.4-beta.9 (beta)

Beta pre-release: offered only to people who turned on Help → Receive beta updates.
Changes since 0.9.4-beta.1.

New: trending effects and transitions (CapCut-style)

• Effects → More effects → Trending: Beat zoom, Flash, Motion trails, Mirror, RGB split (on the beat or steady), Cinematic fade & zoom, Intro move (swipe left/right/up, spin, zoom), Before & after (a white line sweeps across revealing the real picture) and Speed blur (the "velocity" look). Each has its own settings and can be reordered or bypassed in the effect stack and saved in look presets.
• Seven new transitions: Whip pan left and right, Spin, Zoom punch, Glitch, Cinematic zoom fade, and Before & after. Hover a transition to preview it.
• Three new looks: Y2K, Cozy, and Clean & bright, each with its own sliders.
• Speed ramps (velocity) and slow motion for video clips are under Motion → Speed.

Local AI without Docker

• Stable Diffusion and Chatterbox install with one click into SceneForge's own folder. No Docker, no administrator rights. A setup screen shows progress, and you can keep working while it runs.
• Kokoro voices built in (English, Spanish, French, Hindi, Italian, Portuguese), running on your computer after a one-time download of about 120 MB.
• Gemini voices (cloud, free AI Studio key): 30 voices covering 100+ languages, including Arabic.
• "Use my NVIDIA graphics card" is now saved, is chosen automatically when an NVIDIA driver is found, and can be switched later without downloading the models again.
• Engines left running by an earlier SceneForge are replaced automatically. On Windows, engines now close together with SceneForge.
• Repair Chatterbox fixes a Chatterbox that failed to load. The "'NoneType' object is not callable" error is fixed.
• Voicebox (optional): if the Voicebox app is open, its voices appear in SceneForge. Stuck Voicebox jobs are detected and can be cleared.

Editing

• Insert a picture or video between two parts: a marker shows where it will go. Or right-click a part and choose Insert picture or video before… / after….
• Timeline tools now explain what they do. Blade also cuts parts on the Picture track.
• A tip appears when a part has no narration yet. Error messages now include a "What to do" line.
• Captions for songs: much better at catching sung vocals.
• Laptop screens and small windows: tabs, preview and timeline no longer get squeezed or cut off.
• One search-box design everywhere, with a live match count.

Reliability

• The editor stays responsive while an AI engine is busy or not answering.
• Help → Export diagnostics now includes the AI engine logs and setup state.
• Fixed a start-up error in the packaged app that affected installer builds since beta.3.

Notes

• Local AI engines download their models on first use (several GB for Stable Diffusion), so the first start needs an internet connection.
• This is a beta. If something fails, use Help → Export diagnostics and attach the zip to an issue.

SceneForge Studio 0.9.4-beta.1

Draft created by the Release workflow. Publish only after every job is green and the owner approves.

0.9.3

Correct asynchronous API polling in story browser regressions

Run 127 read an intermediate Script-to-Scenes Redo snapshot because waitForFunction treated the async predicate Promise as truthy before inspecting its resolved result. Replace all async API waits in the story harness with bounded Node polling.

Keep exact ordered scene IDs, source-scene preservation and imported narration assertions. Exercise three batch Undo/Redo cycles with 300ms delays on both restored scenes and assert all six restores occur.

Validation: final story harness passed three complete runs (nine delayed batch cycles); full nine-harness browser command passed. Tests and documentation only; runtime and compiled frontend unchanged.

0.9.2

release: SceneForge Studio 0.9.2 — video inside text fits the screen

• render_knockout_card shrinks the font so the widest line (plus outline)
  fits 92% of the frame; long titles at big sizes no longer run off the edges
• panel preview is a true 16:9 frame at real proportions, fitted the same way
• new option: put the person or object in front of the title (runs the
  still-image or video subject cutout after adding the card)
• test_091_fixes.py: shrink-to-fit check

0.9.1

SceneForge Studio 0.9.1

Fixes from the 0.9.0 Windows test, and one new title style. Everything in 0.9.0 is included.

New

• Video inside text (Overlays tab): a solid colour card with your title cut out of it, so the scene's video or picture plays inside the letters, like a documentary "1942" or place-name title. Choose the font, card colour and opacity, letter size, position and an outline.

Fixes

• Subject cutout no longer gets stuck on "Cutting out the subject…". When a video cutout stopped, Windows still held the half-written file, the clean-up crashed, and the job never reported back. FFmpeg is now stopped and waited for before its files are removed, and clean-up never crashes the job.
• Smaller, more useful logs: the hang watchdog no longer dumps every thread's stack for event and media streams, which are long by design (one log had grown to 57,000 lines).
• Help → Collect diagnostics now also saves the backend report (recent render errors, settings without API keys, model status) next to the logs.

Docs

• README screenshots retaken with the 0.9 interface, with new pictures of the Typewriter box and Video inside text.

0.9.0

release: SceneForge Studio 0.9.0 — friendlier editing

Fixes from the 0.8.0 Windows test:
- Beats menu text no longer clipped; track names stay on one line ("Music & SFX" for A3)
- the editor can no longer scroll sideways (left edge was cut off)
- waveforms on A2 clip sound and A3-A8 timeline clips

Usability:
- Typewriter in its own box (switch, speed, timing, sound, volume, preview)
  with GET /api/scenes/{id}/typewriter-preview
- Effects tab (Split toning -> Look presets) sorted into collapsible groups

Roadmap items:
- auto-ducking for A3-A8 clips (audio_clips[].duck, sidechain on scene audio)
- restore points (automatic + named; restore opens a copy)
- "Try a sample project" (generated media, POST /api/sample-project)
- caption translation to English / bilingual with Local Whisper
- title templates (lower third, intro, outro, chapter, quote)

CI: actions v6 (Node 24), checks pinned to ubuntu-24.04, new test_090_features.py.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NxBsZAuJAXTo16eK2ZjqRz

0.7.1

SceneForge Studio 0.7.1

A security and usability update to 0.7.0.

Download

System	File
Windows 10/11 (64-bit)	SceneForge-Studio-0.7.1-Windows-x64-Setup.exe
Linux	.AppImage or .deb
macOS (Apple Silicon, experimental)	.dmg

Existing users are offered this update automatically.

Security

• urllib3 2.8.0 is now bundled. It fixes three high-severity and three moderate-severity advisories.
• The developer build tools move to Vite 7, with patched esbuild; these tools are not part of the installer.
• npm audit reports 0 vulnerabilities.

Usability

• Inspector tabs sit in one tidy row with a More ▾ menu, so they no longer wrap.
• Workspaces: Edit, Color, Audio and Review, under View → Workspace.
• Timeline tracks can be collapsed, and Compact tracks gives a smaller timeline.
• Keyboard shortcuts: press ? for the list, and a short first-run tour (Help → Show tour).
• Help → AI models: see, download or delete the Whisper, background-removal and voice-isolation models.
• Help → Export diagnostics: creates one zip to attach to bug reports. Keys and secrets are never included.

0.5.3

v0.5.3

v0.5.3: version and release notes