Simple to explore.
A lot of work to get there.
A skull you can rotate right away. A skeleton whose parts you can explore. The idea was small. Making it easy to use took many attempts, corrections and decisions.
01Straight to the object
It started with a specific wish: to look at a human skull and skeleton from every angle, without navigating overview pages or learning a complex specialist application first.
“I want high-quality presentation of information rather than generic spam content.”
That is the ambition behind these pages: a question should lead as directly as possible to a visual answer. The object is right there. You can rotate it, move closer and select individual structures.
02Building through conversation
I described the idea, tried drafts, sent back screenshots and pointed out what still did not work. The AI assistant turned that feedback into code, alternatives and corrections. Decisions about appearance, behaviour and priorities came from actually using the result.
The first draft was only a start. Labels were too large or hard to read. Controls took space away from the model. Explanations covered the body. What looked good on a large screen also had to work in a small browser window. Each observation became another specific request for a change.
03Room for the body
The skull needed to be large and central. The canvas was therefore extended across the whole page. Controls and explanations sit over it, instead of shrinking the model into whatever space the layout leaves behind.
The labels went through several revisions too: consistent colours for each structure, softer colours on the model and matching connecting lines. Selecting a part fades its surroundings. Anchors that are out of view should not leave distracting labels on screen.
The skull gained a playful setting inspired by Día de Muertos: paper flags, flowers, small skulls and movement. These decorations were also reduced and rearranged to give the object more room. The skeleton uses a quieter setting.
04The small decisions
Much of the work went into things you ideally barely notice. When should automatic rotation stop? How should labels appear after the opening animation? How compact can the view previews be? Which buttons does this particular object actually need?
The skull also gained a voice. After trying different phrases and comparing voices, I chose OpenAI’s Marin. The short greetings are recordings generated in advance. Jaw motion is prepared too, then played in time with the audio. It is a simplified animation, not an accurate simulation of human speech movements.
05Clear and checkable
The exhibits support twelve languages: German, English, French, Spanish, Arabic, Turkish, Russian, Chinese, Farsi, Italian, Swahili and Tamil. The browser’s language is the starting point; visitors can also choose a language themselves.
The short anatomical descriptions were checked against the cited sources. The skeleton translations received additional checks for terminology, numbering and left/right references. “Left” means the body’s left side, not the left half of the screen.
One example of the precision needed: the skeleton overview initially labelled the head “Frontal bone.” That became a separate selection for the whole skull, while the frontal bone remained available as an individual structure. The correction applies to all twelve languages.
The translations were created and checked with AI assistance. This does not amount to approval by medical specialists or native-speaking editors. The model notes explain further limitations. The models and anatomical references come from the external sources linked in each exhibit; I did not produce new anatomical measurements.
06A shared foundation
The first skull viewer became a shared technical foundation for multiple exhibits. Language selection, rotation, zoom, labels and selection can be reused. Each exhibit supplies its own model, text, colours and available features.
A skeleton therefore does not need a speech button or a teeth toggle just because the skull has them. Separate domains can present different objects without rebuilding every new exhibit from scratch.
07The work behind it
No reliable total of working hours was recorded. The effort is visible in the revisions: visual design, spatial labels, mobile layouts, sound and motion, translations, source notes and finally publishing under separate domains.
My contribution was to look closely, try things and keep describing what needed to become simpler, clearer or better-looking. AI accelerated the implementation. Checking and making decisions remained part of the work.