Skip to content
AI Primer
workflow

Fable 5.2 recreates a Three.js apartment in Blender with Python

MengTo demonstrated Fable 5.2 writing Python to recreate his Three.js apartment in Blender. He reports faster completion than Opus 5.5 at roughly similar cost.

2 min read
Fable 5.2 recreates a Three.js apartment in Blender with Python
Fable 5.2 recreates a Three.js apartment in Blender with Python

TL;DR

  • Fable 5.2 recreated a Three.js apartment in Blender, as MengTo demonstrated in a video tour.
  • The model writes Python, and can use the optional dream-loop skill, according to MengTo.
  • Comparable projects finished faster at roughly the same cost as Opus 5.5, MengTo reported in a follow-up.

The optional dream-loop skill uses a separate AI critic to compare live screenshots against generated target images. Blender's command-line tools can execute Python files and render without opening the interface.

A tutorial room becomes a Blender apartment

MengTo first asked Opus 5.5 to recreate his room in Three.js for a wonderfully specific purpose: animating his tutorial videos. The Fable 5.2 follow-up brings his apartment into Blender.

After sharing the result, he joked in a reply that “we're all pro blender-ers now!”

Python and the dream-loop skill

Fable writes the Python, MengTo confirmed when asked about the workflow. He also named dream-loop as a skill the model can use.

The skill's published workflow has five steps:

  1. Generate a high-quality target screenshot with an image model.
  2. Build the game, app or scene with that target in mind.
  3. Have a separate AI critic compare the live screenshot with the target and provide feedback.
  4. Repeat the build-and-critique cycle until the critic is satisfied.
  5. Optionally generate a better target from the current result and repeat the process.

MengTo presented the skill as an option, without specifying which stages ran on this apartment. No exact build prompt or Python file accompanies these posts; one brief follow-up says only “working on it!”

Faster completion at similar cost

Fable finished projects that Opus 5.5 had struggled with, MengTo said in another reply. Across multiple projects of similar types, he reported faster completion at nearly the same cost.

His original description also called Fable more accurate and token-efficient. Those were his hands-on impressions, with no token totals or timings attached to the comparison.

Runpod and the next renders

MengTo credited discovering Runpod with getting him into Blender. The service offers persistent GPU compute through its Pods product.

He described two further experiments inside the apartment:

Further reading

Discussion across the web

Where this story is being discussed, in original context.

On X· 4 threads
A tutorial room becomes a Blender apartment2 posts
Python and the dream-loop skill1 post
Faster completion at similar cost1 post
Runpod and the next renders2 posts
Share on X