Same prompt, two models: Opus 5.5 vs Fable 5.2 build Blender drone shots from scratch
airusk · reddit · 2026-11-15
The author gave Claude Code an identical one-line Korean prompt to two models — "one-take drone shot over Gwanghwamun Square" and "Yeouido fireworks festival" — each model wrote its own Blender Python script and rendered with EEVEE on the same RTX 3060.
Key observations:
- Opus used no external assets and chose blue hour to make simple buildings hold up; clips are short (22–24s) and freeze at the end
- Fable used just one CC0 sky HDRI but built a far denser city (24,000 buildings, 2,900 trees, 1,500 people); its fireworks shot dives under Wonhyo Bridge and ends with a Niagara curtain
- Both fail at statues, still blocky up close
- Per-frame render time: Opus 0.7–1.1s, Fable 1.9–3.2s
A neat head-to-head revealing very different coding and scene-planning styles.
More from coding & agent
- Why an LLM That Memorized 10 Trillion Tokens of Code Still Acts Like a Junior in Your Repo — RandalSchwartz · 2026-11-15
- autoresearch@home: hundreds of AI agents collaborate on ML research via shared memory — GregoryDiamos · 2026-11-15
- Enterprise AI Needs Tools, Context and Evals, Not Omniscient Models — hwbhatti · 2026-11-15
- Prompt, context, harness, loop engineering are all just "knowing what you want" — mr_samosaman · 2026-11-15
- llm-memory: one SQLite file as shared memory for all your LLMs, via MCP — Mordicus1973 · 2026-11-15
- MCP server gives coding agents a designer: components rebuilt to match your brand — LuckyTNTyt · 2026-11-15