🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦
Build a 3D \"online temple visit\" game with a single Claude AI prompt: the full recipe, shared for free
Quay lại Blog

Build a 3D \"online temple visit\" game with a single Claude AI prompt: the full recipe, shared for free

2 tháng 11, 202611 phút đọcVũ Khôi
English
#English
Vũ Khôi

Vũ Khôi

Tác giả tại Wotech Electronics Blog

A 3D online temple-visit game built with a single Claude AI prompt: a Vietnamese temple at dusk, bronze incense burner, bodhi tree with red ribbons and a chibi pilgrim

Build a 3D "online temple visit" game with a single Claude AI prompt: the full recipe, shared for free

No game-programming skills and no 15–20 USD tool purchase required: a member of Vietnam's AI community has shared, for free, a complete "mega prompt" that makes Claude AI build an entire 3D web game about visiting a temple for blessings — triple gate, incense burner, bodhi tree, incense lighting, bell ringing and fortune sticks included — all inside a single HTML file you just open and play.

Quick answer

  • What you do: copy one long prompt (written like a product spec), paste it into Claude AI, and receive back a single index.html file that runs in any browser.
  • What the author recommends: a Claude Pro plan — ideally Max 5x — so the session does not hit token limits halfway, because both the prompt and the output are very long.
  • The tech: a 3D game using the Three.js library via CDN; every shape is built from code (low-poly), every sound is synthesized with Web Audio — no external images, models or music, and no server needed.
  • The game: an ancient Vietnamese temple at dusk with five mini-activities: lighting incense, ringing the bell, playing the wooden fish drum to a rhythm (scored with a "Peace of mind" score), writing a wish, and drawing a fortune stick for fun.
  • The cost: the prompt is shared with the community for free; the game itself has no login, no data collection, no ads and no payments.

What is an "online temple visit" game?

It is a 3D web game that simulates a blessing visit to a temple: you enter your name, guide a chibi character in a traditional lam outfit around an old Vietnamese temple courtyard, and walk up to activity points to light incense, ring the bell, tap the wooden fish, write a wish that gets hung as a red silk ribbon on the bodhi tree or floated as a lantern on the lotus pond, and draw a New Year fortune stick that is explicitly for entertainment. Everything runs in a phone or desktop browser, portrait-first, with nothing to install.

The remarkable part is not the graphics but the method: the whole game — 3D scene, character, 30–60 AI pilgrims wandering around, progress saving and a share-image feature — is generated from one single prompt, according to its author Nguyen Huy Hung, who posted it in the AI Artworks VN community.

Why is one prompt enough?

Because this prompt is not written as "make me a game". It is written like a miniature product requirements document (PRD):

  • It opens by locking the working style: "Build it completely in one go, do not ask me anything. Where something is unclear, pick the most sensible option yourself and carry on." That one line stops the AI from pausing to ask about every detail — the point where many people give up on "vibe coding".
  • It then splits into clear blocks: the deliverable, the 3D setting, character and controls, the activity list, the crowd, saving and sharing, quality standards, and the output format with a self-check list.

In short, the prompt writer acts as product manager and the AI as the development team: the clearer the spec, the less fixing the result needs.

Anatomy of the mega prompt

Prompt block Key contents What it achieves
The deliverable One single index.html file that opens directly in a browser; Three.js via CDN; Vietnamese UI, mobile-first Locks the output format; no half-finished project, no installs
The 3D setting A Vietnamese temple from sunset into night: triple gate, tiled courtyard, big bronze incense burner, main hall with curved red-tile roof, stylized Buddha statue, bodhi tree, lotus pond, red lanterns, bell tower; temple name sign in Vietnamese script Defines the style and the checklist of objects that must exist
Character and movement A chibi pilgrim in a lam outfit seen from behind; virtual joystick on phones, WASD/arrow keys on desktop; drag to orbit the camera; collision with objects Specifies a playable experience, not just a pretty scene
The activities Five mini-games: light incense, ring the bell, rhythm wooden-fish chanting, write a wish, draw a fortune The gameplay core, each described with interaction and feedback
The AI crowd 30–60 pilgrims built with InstancedMesh, speech bubbles with wishes, groups arriving together, a counter reading "… people came to pray today" A lively atmosphere that stays light on performance
Save and share localStorage: name, incense count, best Peace score, wishes, fortunes; a "My prayer book" page; a Share button exporting a 1080×1350 portrait image via the Web Share API Retention and organic sharing
Quality bar Opens in under 3 seconds, welcome screen and a 3-step tutorial; pixel ratio capped at 2; no new objects created per frame Performance constraints for mid-range phones
Output and self-check Return the full file plus a self-check list: walked the whole courtyard, did all 5 activities, tested sharing, tested portrait frame; anything not done must be stated Forces the AI to audit itself and admit gaps instead of delivering silently

What are the five activities?

  1. Lighting incense: choose 1 or 3 sticks; the character presses palms together, places the incense in the burner, smoke rises and the stick tips glow as they burn down.
  2. Ringing the bell: a light ring shrinks toward its center — tap exactly as it closes to strike the bell; on-beat hits ring out long with a slight screen shake.
  3. Wooden fish chanting: a rhythm game of about 30 seconds; tap the wooden fish on the beat rings and get rated "Centered / OK / Off-beat". At the end you receive a Peace of mind score from 0–100 with a gentle line of praise.
  4. Writing a wish: up to 120 characters for family, health, exams, safety… The wish is written onto a red silk ribbon hung on the bodhi tree, or onto a lotus lantern floated on the pond — your choice. A simple profanity filter is applied.
  5. Drawing a fortune (for fun): shake the fortune-stick tube by shaking the phone or press-and-hold; a stick drops with one of about 40 positive messages. The prompt explicitly requires: no bad predictions, no promises of money or healing, and always a small line reading "For entertainment only".

What do you need before trying it?

  • A paid Claude plan. The author recommends Claude Pro, and suggests Max 5x "so you do not hit the token limit mid-run" — understandable, since a long spec prompt plus a game file of thousands of lines consumes a large amount of tokens in one session.
  • A modern browser on a phone or computer to open the result. No extra software, no server.
  • A few minutes for the checking round: open the game, walk the whole courtyard, try all five activities, test the Share button and the portrait phone frame — exactly the self-check list the prompt asks the AI to report.

Four steps to do it yourself

  1. Get the original prompt from the author's post in the AI Artworks VN group; you can rename the game, the temple and setting details inside the prompt as you like.
  2. Paste the entire prompt into a fresh Claude conversation and send it once; wait for the complete index.html plus the self-check list.
  3. Save the file and open it in a browser. If the self-check lists anything as not done, reply in the same conversation and ask for exactly that item to be completed.
  4. Share the file or put it on any static host if you want friends to play via a link — a single HTML file makes publishing trivial.

Honest limitations

  • Quality depends on the model and plan you use. With the same prompt, results can differ noticeably between Claude tiers; a very long prompt also risks being cut off if the session hits its token limit.
  • "One prompt" is a starting point, not a perfection guarantee. The generated game can still have small bugs in collision, rhythm or layout; the value of the self-check list is knowing exactly what needs a follow-up round.
  • Performance varies by device. Constraints like InstancedMesh crowds and a pixel-ratio cap help mid-range phones run smoothly, but older phones may still stutter in crowded scenes.
  • The religious elements are handled deliberately: the prompt forbids mocking the Buddha statue, monsters, win/lose pressure and rankings, and the fortune feature is always labeled entertainment — keep those guardrails if you customize the prompt.
  • Attribution: the author shared the prompt publicly so the community would not have to buy tools; credit the original share when you remix or repost it.

The author's own upgraded version

According to the author, the version he actually runs "differs a lot" from the prompt-generated one: its 3D characters are made separately with the Tripo tool and imported, it adds elements such as hand-seal gestures, it ties into his "Dao Su Tru Ta" (Exorcist Master) game universe, and it is heavily optimized to reduce lag. Readers can try the upgraded build at the address the author announced: trangsan.com/zuzai. It illustrates a realistic path for many AI products: a prompt produces the first playable version, then a human upgrades the parts the AI could not reach.

Frequently asked questions

Can I do this without knowing how to code?

Yes, at the follow-the-guide level: your job is copying the prompt, pasting it into Claude and opening the result file. Being able to read an error message and describe what looks wrong, so the AI can fix it, improves results — but writing code yourself is not required.

Can I use a free Claude account?

Both the prompt and the output are very long, so the author's recommendation is Pro or above (Max 5x ideally) to avoid being stopped by token limits. A free account can try a shortened prompt, but should not expect the full game in one session.

Does the game run on phones?

Yes — the prompt demands a portrait-phone-first design: a virtual joystick, large action buttons, a pixel-ratio cap and InstancedMesh crowds to stay light. Desktops work too, with WASD or arrow keys.

Does the game collect player data?

No. The prompt is explicit: no login, no data collection, no ads, no payments; progress is stored only on the player's own device via localStorage.

Does this replace actually visiting a temple?

No, and the prompt never claims it does: this is an entertainment product — a gentle experience inspired by the culture of temple blessing visits. The fortune feature always carries a "For entertainment only" note and promises neither spiritual efficacy, money nor healing.

Conclusion

The real story here is not a 3D temple game; it is a new way of working with AI: instead of "give an order and hope", write the prompt like a spec — lock the output format, list the components, set the quality bar and make the AI self-check before delivering. Written that way, one prompt can buy you a genuinely playable product, and this free, familiar online-temple mega prompt is an example anyone can try today.

Source: shared by Nguyen Huy Hung in the AI Artworks VN Facebook group (2 Nov 2026), edited and structured by Wotech Electronics.