Changelog

Fixes

• Film to Story analysis now includes CUDA kernels for RTX 50 series GPUs, including RTX 5060 and RTX 5090.
• The app checks real GPU execution before analysis and switches visual descriptions and footage screening to CPU if GPU processing fails.
• The packaged installer is tested with real transcription and visual analysis models on both GPU and CPU.

Fixes

• Bundled the missing SciPy native module required when visual shot descriptions begin after transcription.
• Added a packaged-backend dependency check to catch this failure before future installers are built.
• The sidebar now shows the installed app version.

Improvements

• Footage review has faster shot selection and bulk decisions; generated scripts account for the chosen timing settings.
• Voiceover and render checks handle shot coverage more clearly.
• Gemini and ChatGPT thumbnail generation handle reference attachments more reliably, with more text colours.

Faceless videos and story controls

• Choose Faceless video or With presenter. Faceless videos use a project voice, B-roll and optional source footage. Choose from 28 topic categories, with stronger opening hooks and title guidance.
• AI video B-roll fills the 9:16 Reel frame. Hook-title duration and four styles are available across AI Video Creator, Film to Story and Video to Documentary, with a separate background-music title.
• Film to Story offers 13 narrative styles with descriptions and film-character reference images for thumbnails, including Viral Fiction.

New tool

• AI Video Creator — a saved character (AI picture or photo, motion clips, default voice), a presenter share target, a script that alternates presenter segments and B-roll, the presenter animated on your PC with free bundled models (LivePortrait motion + MuseTalk lips), AI-placed camera zooms, Reel and YouTube renders without any film.
• Characters page: reference pictures from Gemini / ChatGPT / Grok with attachments, motion library (record yourself or Grok Imagine clips), one character across projects.

B-roll

• AI video clips: Grok Imagine text→video from the script's prompt, right on the B-roll card (6 or 10 s, cut exactly to the beat at render, adjustable start).
• Real footage requests: the script asks for a specific clip, you import it.
• Grok added as a third picture provider on every B-roll row.

Render

• The Render page lists presenter segments that are missing or changed and renders them in one go before the final video.

Fixed

• Video → Documentary analysis stopped at the descriptions stage in the installed app (transformers … __init__.pyc) — fixed in the bundled backend.
• The Read on-screen text (OCR) switch could not be turned on in the installed app (the models were shipped but not found); when OCR is really missing the switch now says so.
• Sign in with Google inside the ChatGPT / Grok login windows no longer hits “This browser or app may not be secure” — the same fix Gemini already had.

Installer

• Setup now installs for all users (one Windows prompt at the start) so any folder works, including a second drive.
• The 3.9 GB download shows real progress: size, percent, speed and time left; it resumes after a network drop.

AI

• Grok can be connected in Integrations → AI Chat and used to generate pictures next to Gemini and ChatGPT.

New tools

• Film To Story — import (PC, link, YouTube page, magnet), shot map with credits filter and optional OCR, story styles (Faithful, Fast & dramatic, Twist first, Villain's POV, Mystery, Viral fiction with genre picker), Master Prompt, strict story validation, per-beat voiceover (ElevenLabs / Gradium), music with ducking, Reel 9:16 and YouTube 16:9 renders, copyright shield, captions, hook title, per-format thumbnails / covers with composed text, publish modal, cloud snapshot + restore on another PC, TMDB posters.
• Video → Documentary — 2–8 sources in any language, per-source analysis merged into one shot map, OCR facts (Latin + Arabic), presenter shots hidden, channel logo detection (zoom out / blur / mark), original fact-based script with references and B-roll pictures (charts drawn locally, maps / diagrams / illustrations generated by AI in a chosen style), script QA with "ask the AI to fix this beat", rhythm marks (pause, fade to black, echo→black on the last word, slow motion) with Render-page switches and a fade-length slider.
• Video → Facebook Post, Facebook Post Rewriter, Article → Facebook Post, Blogger Article → Facebook Post (funnel: teaser + first-comment link, auto or via Telegram).
• Video → Blog Article — SEO article with sections, FAQ, meta, labels, per-section pictures hosted on ImgBB, publish live or as a draft to Blogger.

AI

• Hidden-browser generation with your connected Gemini / ChatGPT for clips, scripts and pictures; several projects at once, queueing, human-check pause, provider fallback.
• Pictures for posts, articles, thumbnails and B-roll now come from Gemini / ChatGPT — the FluxAPI integration was removed.
• Master Prompts list the project files an AI IDE may open to verify shots.

App

• Automatic updates: 3.0.1 ships as an installer; from now on new versions download in the background and install when you restart (Settings → About & updates). Users of the old 2.0.1 zip must download 3.0.1 once.
• New Home (hero + one card per tool), My projects grouped by tool with search and sort, every tool has its own page with a New project button.
• Per-tool project folders (video-to-shorts, film-story, video-to-documentary, facebook-posts, blog-articles); existing projects are never moved.
• Unified rendered-video row everywhere: render again, Download with progress, Copy link, cloud state, Publish.
• How it works rewritten for every tool with the same structure; Home cards link to their chapter.
• Interface in English, French and Arabic (RTL).

Voice & integrations

• ElevenLabs cloud key pool (rotation by credits, monthly reset awareness), Gradium local vault with 1,500-character splitting and word timings.
• Integrations page: AI Chat, ElevenLabs, Gradium, ImgBB, Telegram.

Cloud

• Film / Documentary snapshots (state + bundle) synced after every change; restore on another PC with paths relocated; film re-download.
• Thumbnails attached to cloud clips and used by YouTube, Facebook and Instagram publishers.

Added

• Buffer Hub (publish and schedule through Buffer channels, status sync).
• Custom output location and Storage panel in Settings.
• Runtime CPU fallback when the GPU kernel is incompatible.
• Auto-update through GitHub releases.

Fixed

• Parallel-upload logout after Render all (single-flight token refresh).
• Reopened projects show Rendered · Re-render + Retry upload instead of a bare Render.
• Cold-open teaser dropped when toggled after paste.

Added

• Local desktop app (Electron + React) with a Python backend: faster-whisper transcription, Master Prompt generator, filler / pause removal, YOLOv8 face tracking, word captions, NVENC rendering.
• Cloud companion: 24-hour clip bridge, OAuth publishing to YouTube, TikTok, Facebook, Instagram.
• Landing page, privacy policy and terms.