← Back to selected work
AUTOMATION · VIDEO PIPELINE · PRODUCT ENGINEERING
Project Forge
A reproducible Python and FFmpeg pipeline that turns one selected sports training clip into branded, platform-ready 16:9, 9:16, and 1:1 MP4/GIF deliverables.
THE PROBLEM
01One training clip often becomes several repetitive editing jobs.
Sports coaches, studios, and creators reuse the same footage across YouTube, Reels, Shorts, websites, and social feeds. Manually rebuilding each composition wastes time and makes branding inconsistent.
Repetitive production
The same title, logo, captions, crop, and export settings are rebuilt for every platform.
Unsafe resizing
Simple center crops can cut off the athlete, while generic blurred backgrounds often look unintentional.
Inconsistent delivery
Different dimensions, codecs, frame rates, and naming conventions create avoidable review work.
THE SOLUTION
02Automate the repeatable work without pretending every video needs AI.
Project Forge accepts a deliberately selected clip and a JSON configuration, then builds format-specific layouts with a stable subject anchor, branded overlays, burned-in captions, and verified exports.
WSL / UBUNTUREPRODUCIBLE RUN
python pipeline.py --config config.json --clean
Landscape
1920 × 1080 full-width output for YouTube, websites, presentations, and horizontal campaigns.
Portrait
1080 × 1920 card composition designed for Reels, Shorts, TikTok, and Stories.
Square
1080 × 1080 feed-ready composition with a dedicated subtitle-safe region.
ARCHITECTURE
03A narrow, testable rendering pipeline.
- 01
Load selected source and JSON config
Input paths, project metadata, title copy, subject anchor, and output specifications stay outside the rendering code.
- 02
Generate format-specific visual assets
Python and Pillow create title overlays while SRT captions are converted to ASS with per-format positioning and typography.
- 03
Compose each layout with FFmpeg
The landscape, portrait, and square outputs use separate render graphs rather than applying one generic resize.
- 04
Export production and preview formats
Each layout produces an H.264 MP4 and a lightweight GIF preview with deterministic file naming.
- 05
Verify the deliverables
FFprobe checks resolution, frame rate, codec, and sample aspect ratio; GitHub Actions repeats the smoke test on pushes and pull requests.
Python 3.10+FFmpegFFprobePillowASS subtitlesJSON configGitHub Actions
PRODUCT DECISIONS
04Stability before speculative intelligence.
BUILT IN V1
Static-anchor reframing
A normalized focus point creates predictable crops for a selected clip and keeps the workflow easy to rerun.
BUILT IN V1
Format-specific compositions
Portrait and square outputs are designed as layouts, not merely resized versions of the landscape frame.
EXPLICITLY OUT OF SCOPE
No automatic person tracking
V1 does not claim per-frame detection, multi-person tracking, primary-subject selection, or AI-generated camera motion.
WHY
A commercial promise the system can keep
The product demonstrates reliable batch formatting for suitable clips instead of promising zero-configuration support for arbitrary footage.
RESULTS
05A complete v1 release, not only three exported videos.
OutputSpecificationStatus
16:9 MP4 + GIF1920 × 1080 · 30 fps · H.264PASS
9:16 MP4 + GIF1080 × 1920 · 30 fps · H.264PASS
1:1 MP4 + GIF1080 × 1080 · 30 fps · H.264PASS
Automated verificationDimensions · codec · fps · SARPASS
CI smoke testFull render and validation on GitHub ActionsPASS
One selected input clip. One command. Six branded deliverables.
CURRENT STATUS
06V1 is complete and frozen; active commercial development is paused.
ENGINEERING STATUS
Complete and reproducible
The multi-format renderer, branding configuration, preview generation, output verification, CI smoke test, repository, and case study are finished.
MARKET LEARNING
Resize alone is not the product
Single-clip formatting is vulnerable to price pressure and does not prove creative editing, hook development, sound design, or campaign strategy.
ACTIVE HYPOTHESIS
High-volume media operations, only with evidence
A future version would focus on high-volume language, platform, or brand operations—but only after a buyer presents real volume, budget, and workflow pain.
DEVELOPMENT GATE
No v1.1 without buying evidence
A larger batch system will only be developed after a paid pilot, a formal quote request, or strong evidence that teams already pay for this workflow.
V1 COMPLETE · COMMERCIAL DEVELOPMENT PAUSED
The engineering proof is finished, and speculative expansion has stopped.
Project Forge remains a public engineering case study. It will only be reactivated when a real buyer demonstrates sufficient volume, budget, and operational need.