Genesis Artvision Engine

By ANSNEW TECH

Offline procedural video factory. Press GENERATE VIDEO and the app invents a seed, engine, style, palette, background, layout, motion, voice, and soundtrack, then encodes an MP4 locally with FFmpeg.

No stock footage, prompt, or music library is required. Trending Brief reads public news RSS when the machine is online. Optional OpenRouter advice can polish that brief or suggest creative direction. Frames, pictures, voice, and audio stay on this machine.

What it makes

Three engines. The engine owns the concept. The style owns grade, chrome, and motion. Each new seed also picks a procedural background and a safe layout. Those names are stored in the project spec so History can reprint the same picture.

Engine	Style	What you get
kids_storybook — Kids Storybook	storybook	Picture-book pages, a drawn noun, and a slow kids voice
how_it_works — How It Works	classroom	Classroom diagrams plus a calm narrator
trend_brief — Trending Brief	pulse	A current RSS headline, a short spoken brief, kinetic ticker

Leave Engine and Style on Random and they pair together. Each seed still varies the backdrop and layout (desk, window, chalkboard, aurora, radar, mirrored cards, and so on).

Requirements

• Python 3.11+ (tested on 3.14)
• FFmpeg on PATH
• Windows first (Linux/macOS should work with FFmpeg; kids voice prefers Windows SAPI)

Installation

cd random_art_video_factory
python -m pip install -r requirements.txt

Install FFmpeg if needed:

• Windows: winget install Gyan.FFmpeg
• Or download from https://ffmpeg.org and add bin to PATH

Verify:

ffmpeg -version
python main.py --test

Run GUI

python main.py

The window is a two-pane studio: settings across the top, Activity beside Preview, transport and quick actions always visible.

• Edit preset: Draft, Standard, or Master
• Captions: SRT sidecar, burned, both, or off
• New prompt: Ctrl+P, then Ctrl+Enter to submit
• Shortcuts: Ctrl+Enter generate, Esc stop, Ctrl+Shift+P pause, Ctrl+Shift+R resume, Ctrl+H history, Ctrl+O output

When a render finishes, the result card shows the thumbnail, seed, local path, captions, manifest, and either a youtu.be link or a short YouTube error.

Run CLI

python main.py --generate
python main.py --generate --count 10
python main.py --generate --duration 30 --resolution 1920x1080 --fps 30
python main.py --generate --seed 847293847
python main.py --generate --engine kids_storybook --style storybook
python main.py --generate --engine how_it_works --style classroom
python main.py --generate --engine trend_brief --style pulse
python main.py --test
python main.py --prompt "A bedtime story about a brave orange cat who finds the moon"
python main.py --prompt "Explain how rain is made" --ai
python main.py --generate --edit-preset master --captions both --edit-intensity 1.15
python main.py --generate --youtube --engine how_it_works
python main.py --generate --no-audio --test

Live Trending Brief

When you pick Trending Brief and the machine is online, the factory reads public RSS feeds (Google News, Google Trends, BBC, NPR), picks one current headline, and turns it into a four-beat brief: HOOK, WHY, CATCH, NEXT. It uses titles and descriptions from the feed only — it does not scrape article pages.

If OPENROUTER_API_KEY is set, OpenRouter only rewrites those facts into easy spoken lines. Without a key, a local extractive brief is used. If the web is down, the evergreen catalog is used. Set trend_feed.enabled: false in config.yaml to skip live news.

A new generate with the same seed on another day can pick a different story because the news moved on. History reprint uses the stored brief and does not refetch.

Seeds and replay

Every project gets an integer seed. The same seed with the same resolution, fps, and duration reproduces the same video, including the stored background and layout variants.

History stores the complete specification. Make again from seed replays those stored parameters instead of the current GUI controls. Older projects without variant fields keep the original desk, whiteboard, or neon look.

Delivery files

Each successful render can write a matched set in output/:

• .mp4 — H.264/AAC video
• .srt — narration-synchronized captions
• .json — spec, editorial plan, AI hash, and QC report
• .jpg — thumbnail when enabled

QC checks muxed streams, resolution, FPS, frame count, A/V drift, caption timing, loudness, silence, clipping, and sampled black or frozen frames. A broken soundtrack stops the export.

YouTube upload

The factory can upload the finished MP4 to your channel with an SEO title, hashtags, and a 1280×720 thumbnail. This uses Google’s official OAuth + YouTube Data API.

1. Enable YouTube Data API v3 and create a Desktop OAuth client.
2. Save data/youtube/client_secret.json (copy from data/youtube/oauth_client.example.json) or set YOUTUBE_CLIENT_ID / YOUTUBE_CLIENT_SECRET in .env.
3. In the GUI: YouTube → Connect / switch channel… and pick the Brand Account.
4. Check Upload to YouTube. Default privacy is Unlisted.

Quota is about 6 uploads per day (youtube.daily_limit in config.yaml). Kids Storybook is marked Made for Kids.

Optional AI advisor

Suggestions only. Rendering stays local.

copy .env.example .env

OPENROUTER_API_KEY=sk-or-v1-your-key-here

python main.py --curate
python main.py --generate --ai --engine kids_storybook --count 1

Or enable AI advisor in the GUI. Defaults keep ai.enabled: false.

Project structure

random_art_video_factory/
  main.py
  config.yaml
  requirements.txt
  app/
    art/          # engines, layouts, backgrounds, typography
    audio/        # voice, music, mastering
    ai/           # optional OpenRouter advisor
    core/         # randomizer, factory, scheduler
    database/     # SQLite history
    gui/          # PySide6 studio
    publish/      # YouTube upload and SEO thumb
    video/        # FFmpeg, captions, QC
    utils/
  assets/         # app icons only
  data/youtube/   # OAuth example; tokens stay local
  output/         # generated MP4 + sidecars
  temp/
  tests/

Word cards and AI scenes are drawn on demand and cached locally. They are not shipped in the repo.

Tests

python -m pytest tests/ -q
python main.py --test

Troubleshooting

Problem	Fix
FFmpeg was not found	Install FFmpeg and restart the terminal
GUI will not start	pip install PySide6
Slow renders	Use --test, Draft, or a shorter duration
Soundtrack generation failed	Check the log and TTS/FFmpeg; export stops so a silent file is not treated as finished
Disk filling up	File → Clean temp files or delete temp/
Reproduce a video	Use the seed from History
YouTube 403	Enable YouTube Data API v3 on the same Cloud project as the OAuth client
Wrong YouTube channel	YouTube → Connect / switch channel… and pick the Brand Account

---

© ANSNEW TECH — Genesis Artvision Engine