INDEX / 02 — PROJECT
Object Worlds
02
STATUS
LIVE · AI VIDEO PIPELINE · PYTHON · CLAUDE API · FLUX · WAN 2.5
03
PITCH
“One idea. One command. One finished video. No human in the loop.”
04
STATS
PIPELINE STEPS
12
COST PER VIDEO
$1.40
VIDEO LENGTH
22.5s
SCENE COUNT
3
MODELS USED
5
05
DESCRIPTION
Full automated AI video pipeline. You write an idea in Russian into idea.txt. One Python command launches everything:
- Idea validation — Sonnet scores conflict potential 1–10 in 1 second.
- Claude Opus expands idea into: 3 scene prompts for FLUX (hook → reveal → resolution order), 3 narration lines (max 5 words each), 1 hook line (provocative, max 8 words), 1 YouTube title with emoji.
- Sonnet generates description — episode number, 2 lines, CTA, 5 hashtags.
- FLUX 1.1 Pro generates 3 images at 9:16 aspect ratio.
- Sonnet QC checks each image — score 1–10, passes if ≥7.
- Wan 2.7 i2v generates 3 videos (7 seconds each, 1080p) from images.
- FFmpeg concat — joins 3 videos in Claude's specified order.
- Black hook — 1.5s black screen with hook text (Orbitron font) + random SFX.
- FFmpeg assembly — adds subtitles, watermark, CTA overlay, background music.
- ElevenLabs TTS — generates 3 voiceover MP3s (female, multilingual v2).
- FFmpeg voiceover mix — overlays voice, ducks music to 50% during speech.
- Final output: 22.5-second short ready to upload.
Resume support. If the pipeline crashes at any step, run_status.json remembers where it stopped. Restart continues from that point.
06
SHOWCASE
07