Castle Bake-off

Castle Bake-off is a visual comparison table disguised as a medieval field trip. Eight named models tackle the same castle-world specification in low-poly, semi-realistic, and very realistic styles, producing 24 self-contained Three.js scenes to browse, open, and pair off.

What is Castle Bake-off?

The landing page keeps the experiment legible: static cards first, full WebGL worlds only when you ask for one. That makes it easy to scan the matrix by model or art direction before opening a castle, and Compare mode lets two choices share the stage. The fun lies in recurring ingredients interpreted eight ways—towers, walls, terrain, lighting, and whatever personality the generated code manages to smuggle into the same specification. Each world is presented as a single procedural HTML file with no external assets, which gives the collection an appealing ship-in-a-bottle constraint.

What you can do there

• Browse 24 castle-world screenshots
• Open individual procedural Three.js worlds
• Compare two model and art-direction combinations side by side

Why we picked it

Most model comparisons end as scoreboards or walls of text. This one gives the outputs somewhere to live. The fixed brief and three repeated art directions make differences easy to spot, while the openable worlds turn comparison into wandering rather than spreadsheet duty.

How to get the most from it

Scan Gallery mode by style first, then choose two castles that diverge in an interesting way and place them in Compare. Open the full worlds only after narrowing the field; the static cards avoid loading WebGL for all 24 scenes at once. Try holding the model constant while changing style, then reverse the experiment.

Good to know

Treat this as a visual bake-off, not a controlled capability ranking: the exact prompts, settings, selection process, and code provenance are not published on the landing page. The creator is not identified there, and current mobile controls, accessibility, privacy behavior, and individual-world performance remain unconfirmed.