Project case study
Autonomous Arena Studio
A synthetic-fiction pipeline that auto-generates and renders vertical Blender fight videos between original persona fighters.
Problem
Short-form narrative video usually means either live-action production or hours of manual 3D animation per clip. Autonomous Arena Studio automates that loop instead for one constrained fiction: original internet-culture-style persona fighters, a 32-entry bracket, stat-plus- controlled-upset match resolution, and rendered vertical fight videos — with an explicit rule against depicting real people, real events, or copied IP.
Approach
A daily-cycle job seeds a season, resolves matches, and renders each one through local Blender
automation with no paid video credits required by default, producing a full episode package:
video, thumbnail, caption, metadata, a scene manifest, and labeled contact sheets for visual QA.
The default blender_2d mode does deterministic vertical cutout-style animation via the companion
BlendAgent library, assembled to MP4 with ffmpeg, and falls back to a local puppet renderer if
Blender isn't available. An experimental 3D path and an opt-in fal AI video-generation mode also
exist, both gated behind explicit flags. Publishing automation is deliberately left for later.
Outcome
The repo is private. Development moved fast for one concentrated stretch — the commit log runs from initial scaffold through the 2D-renderer pivot and choreography QA entirely within 2026-06-06 and 2026-06-07 — but nothing has landed since; it's been paused for about seven weeks as of this writing.