ShortsGenerator

Automate YouTube Shorts creation locally — script generation, stock video search, TTS voiceover, subtitles, background music, and social media scheduling.

Features

• AI Script Generation — uses g4f (free) or Gemini to generate video scripts
• Stock Video Search — auto-downloads matching clips from Pexels
• Multi-Voice TTS — Supertonic (local, 10 voices, 33 languages), TikTok TTS (fallback), KittenTTS, Qwen3-TTS (local, 9 preset timbres + Voice Design + Voice Clone + character voices 🐶 EL-PERRO / EL-GANCHO / LA-FIERA / SOMBRA, hook/retention steering presets)
• Subtitle Templates — 10 presets (classic, modern_glow, bold_outline, minimal, cinematic, neon, social_viral, floating, news_ticker, karaoke_highlight)
• Background Music — auto-mix from your music library or extract from a video
• Image Stitching — upload multiple images to stitch at the start of the video with configurable duration per image (default 5s)
• Frame Extraction — extract a frame from any generated video for thumbnails
• Aspect Ratios — 9:16, 16:9, 1:1, 4:5, 21:9
• Multi-Business MagicSync — configure multiple API keys for different businesses; select which one to use when scheduling
• Social Media Scheduling — schedule posts to Instagram, TikTok, Facebook, LinkedIn, YouTube via MagicSync
• Hardware Acceleration — auto-detects GPU for faster ffmpeg encoding

YouTube or click on the image.

Local usage:

git clone https://github.com/leamsigc/ShortsGenerator.git
cd ShortsGenerator
cp .env.example .env
#Start the backend
python Backend/main.py

#Start the frontend
cd UI && pnpm i && npx  nuxt dev --tunnel #Tunnel to add the connection to upload the videos to magicsync  

Workflow

Quick Start (Docker)

git clone https://github.com/leamsigc/ShortsGenerator.git
cd ShortsGenerator
cp .env.example .env
# Edit .env — at minimum set PEXELS_API_KEY and IMAGEMAGICK_BINARY
docker compose up -d

Open http://localhost:5000 or http://localhost:3000 for the frontend.

Manual Setup

Backend (Python Flask on :8080)

conda activate shortsgenerator   # required: Python 3.11 env (see onBoard.md).
                                 # main.py exits immediately on any other version.
pip install -r requirements.lock # pre-resolved; plain requirements.txt can stall pip's resolver
cp .env.example .env
# Fill in PEXELS_API_KEY, IMAGEMAGICK_BINARY, etc.
cd Backend
python main.py

Frontend (Nuxt 3 on :3000)

cd UI
npm install
npm run dev

The frontend depends on the backend running on port 8080.

Environment Variables

Variable	Required	Description
PEXELS_API_KEY	Yes	Stock video search
IMAGEMAGICK_BINARY	Yes	Path to ImageMagick convert (e.g. /usr/bin/convert)
TIKTOK_SESSION_ID	No	TikTok TTS fallback
GOOGLE_API_KEY	No	Gemini AI model
ASSEMBLY_AI_API_KEY	No	Cloud subtitle generation
OPENAI_API_KEY	No	OpenAI models
MAGICSYNC_BASE_URL	No	Default MagicSync server URL
MAGICSYNC_API_TOKEN	No	Default MagicSync API token

See EnvironmentVariables.md for the full list.

Usage

1. Generate a script — enter a video subject, select a script template, click Generate
2. Review & edit — modify the script, add search keywords, select subtitle template
3. Upload images (optional) — upload images that will be stitched at the start of the video; set duration per image
4. Select voice — choose TTS engine and voice style (Qwen3 clones/designs can be saved as named "My voices" in your browser and reused as presets)
5. Set aspect ratio — 9:16 (default), 16:9, 1:1, or 4:5
6. Generate video — downloads stock clips, generates TTS, combines everything with subtitles
7. Add music — pick from your music library or extract audio from a video
8. Schedule — in the Videos page, click "Schedule Upload", select a business and platforms, set date/time, and post via MagicSync

Image Stitching

In the Generate page, click the Images section and upload one or more images. Each image becomes a video segment at the start of the final video. Adjust the global duration (default 5s) or set per-image duration. Images are scaled and cropped to match the selected aspect ratio.

Multi-Business MagicSync

In Settings → MagicSync Integration, add API keys for each business. Each business has its own MagicSync URL, API token, and video base URL. When scheduling a video, select which business to use — the correct credentials are sent automatically.

Thumbnail Extraction

POST /api/extract-frame extracts a frame from any generated video at a given timestamp. Use this to generate custom thumbnails.

API Endpoints

Method	Endpoint	Description
POST	/api/generate	Full video generation pipeline
POST	/api/script	Generate script only
POST	/api/search-and-download	Search + download + TTS + combine
POST	/api/regenerate-video	Re-render audio + video only (no AI calls, reuses metadata)
GET	/api/tts/status	TTS engine health (supertonic/tiktok/qwen3)
GET	/api/tts/voices?engine=	Voices + extras per engine (languages, modes, presets)
POST	/api/tts/qwen/preview	Audition a Qwen3 voice/mode without full generation
POST	/api/tts/qwen/clone-reference	Upload a reference clip for Qwen3 voice clone
POST	/api/cancel	Cancel generation
POST	/api/addAudio	Add background music to video
POST	/api/upload-music	Upload music file
POST	/api/download-music-url	Download & extract audio from URL
POST	/api/upload-video	Upload video files
POST	/api/upload-image	Upload images for video stitching
POST	/api/extract-frame	Extract a frame from a video
GET	/api/getVideos	List generated videos
GET	/api/getSongs	List available music
GET	/api/models	List TTS voices
GET	/api/settings	Get global settings
POST	/api/settings	Update global settings
POST	/api/magicsync/accounts	List MagicSync accounts
POST	/api/schedule-to-magicsync	Schedule a post
GET	/api/video/<filename>	Serve generated video
GET	/static/generated_videos/<file>	Static video files

Directory Structure

Backend/
├── main.py              # Flask app, all API routes
├── video.py             # Video processing, ffmpeg, subtitles, images
├── settings.py          # Global defaults
├── gpt.py               # AI script generation
├── search.py            # Pexels stock video search
├── tiktokvoice.py       # TikTok TTS
├── supertonic_tts.py    # Supertonic local TTS
├── qwen3_tts.py         # Qwen3 TTS (preset timbres + Voice Design + Voice Clone)
├── requirements.lock    # Pre-resolved pins (use this for pip install; see Notes)
├── classes/
│   └── Shorts.py        # Core pipeline orchestrator
UI/
├── pages/
│   ├── generate/index.vue  # Video generation workspace
│   ├── videos/index.vue    # Gallery + scheduling
│   └── settings.vue        # Global settings + MagicSync
├── composables/
│   ├── useVideoSettings.ts # Video generation state
│   └── useGlobalSettings.ts # App-wide settings
└── stores/

Notes

• Python 3.11 only. The backend exits on startup with any other version (conda activate shortsgenerator first, in the same terminal). Running it under base (3.10) fails with cryptic native-lib errors (e.g. torchaudio .so load failures) — never install project packages into base.
• Install via the lockfile. pip install -r requirements.lock instead of requirements.txt: the graph (torch/CUDA + qwen-tts → pinned transformers==4.57.3 → huggingface-hub<1.0) makes pip's resolver stall (resolution-too-deep). After changing pins, regenerate with: uv pip compile requirements.txt --python $(which python) -o requirements.lock
• Qwen3-TTS is optional but local. Select it in Settings → TTS Engine for preset timbres, free-form Voice Design, or Voice Clone, plus a preview audition button. Missing GPU deps degrade gracefully to "unavailable" with Supertonic/TikTok fallback. First preview downloads ~3GB of weights once.
• g4f Gemini cookies. The free g4f path reads Firefox/Chrome login cookies via browser-cookie3 (required package). If Settings reports "cookies outdated" with 0 cookies found, check the package is installed and restart the backend — re-logging in can't help until the reader exists. Snap-Firefox profiles are supported. Escape hatches: toggle OFF "Use browser cookies" (cookie-free providers) or set provider gemini with a free GOOGLE_API_KEY.

Contributing

Pull requests are welcome. For major changes, please open an issue first.

License

See LICENSE for details.