Specify editability
State which objects must remain separate, named and editable; otherwise a visually convincing shell may still be unusable.
Explore Interactive 3D robotic hand piano demonstration: a GPT-6 Astra 3D example featuring Three.js, TypeScript. Preview the result and adapt the prompt for your project.
One character, two real workflows. Results vary by task.
Give your next scene a character worth a closer look. Create detailed, textured 3D assets from an image or a few words.
Generate 3D Assets FreeBuild a complete browser-based demonstration of a detailed five-finger robotic hand playing a miniature piano. The visible finger motion, physical key travel, generated notes, and musical timing must be causally connected. Deliver a visually polished, interactive application within the evaluator's time budget. 1. EXPERIENCE: Use a full-screen 3D scene with a precisely modeled robotic hand, articulated fingers, visible wrist mechanisms, and a 25-key keyboard spanning MIDI notes 60 through 84. Show realistic black-key and white-key geometry, independent key movement, fingertip pads, and refined materials. Include overhead, performer-side, and fingertip close-up cameras. Provide synchronized audio after the user activates playback. 2. COMMON MUSICAL INPUT: Use MIDI note numbers as the source of truth. At 96 BPM, play these events, expressed as (start beat, note, duration in beats): (0,60,0.4), (0.5,64,0.4), (1,67,0.4), (1.5,64,0.4), (2,62,0.4), (2.5,65,0.4), (3,69,0.4), (3.5,65,0.4), (4,60,0.4), (4.5,60,0.4), (5,60,1), (5,64,1), (5,67,1). The last three events form a simultaneous chord. Also support standard MIDI file import using an established parser. 3. HAND CONTROL: Model independently articulated fingers and a movable wrist. Plan reachable finger assignments, approach motions, presses, holds, releases, repeated-note articulation, and chord execution. Fingers must contact the correct keys without intersecting neighboring keys or making implausible jumps. Use inverse kinematics and joint limits. Display planned finger assignments and allow manual inspection of individual motions. 4. SOUND CAUSALITY: Generate note-on events only when the corresponding visible key crosses a documented depression threshold because of finger contact. Generate note-off on release, with hysteresis to prevent chatter. MIDI events are planning targets, not an independent audio playback track. A geometric contact-driven key mechanism is acceptable if explicitly identified; full contact dynamics may be used instead. Keys must not move merely because a MIDI event is scheduled. 5. TIMING: Use a consistent musical clock and timestamp actual key-trigger events against target events. Account for audio scheduling and render timing. Expose tempo, transpose, play, pause, restart, loop, and slow-motion inspection. Pausing or restarting must release active notes appropriately. Slowing playback must preserve synchronization between fingers, keys, and audio. 6. DIAGNOSTICS: Display target notes, planned fingers, actual triggered notes, and onset timing errors on an aligned timeline. Report missed notes, extra notes, wrong pitches, repeated-note failures, and stuck notes. Provide a contact inspection overlay showing which fingertip is depressing each key. Record the evidence needed to distinguish successful planning from approximate hand animation. 7. VERIFICATION: Evaluate melody, repeated notes, and the final chord separately. Aim for no wrong or missing notes, a 95th-percentile onset error below 50 ms, and final-chord onset spread below 50 ms. Report actual measurements even when targets are missed. Provide a test that disables finger actuation: the score may continue advancing, but unpressed keys must not generate notes. 8. DELIVERY: Use Three.js, TypeScript, appropriate audio APIs, and established parsing or numerical libraries. Deliver the running application, source code, reproducible musical fixture, assets or generation scripts, and startup instructions. Verify browser audio, controls, camera views, desktop and mobile layout, and repeated playback. All displayed performance claims must come from measured behavior.
Prompt breakdown
Explore “Interactive 3D robotic hand piano demonstration”, a GPT-6 Astra 3D example about Three.js, TypeScript, 3D scene. Check the evidence label to see whether the text is an original prompt or a brief adapted from the public source.
Keep the core outcome, then specify the camera, controls, lighting and what a successful result should do. Test that first before adding more visual detail.
State which objects must remain separate, named and editable; otherwise a visually convincing shell may still be unusable.
Ask the model to run the scene, inspect visible failures and repeat until concrete frame-rate, interaction and layout checks pass.
For large worlds, split the work into districts, assemblies or milestones and approve each section before expanding scope.
Your prompt builds the game or scene. Bring it to life with more detailed characters and props, created in Tripo to replace simple placeholder models.
A reusable task brief reconstructed from the author’s public project description. It is not presented as a verbatim hidden prompt.
Build a kaiju-inspired Three.js game using generated creature models and sound effects. Create readable giant-scale combat and an environment that communicates the size of the creatures.
Set up expression variants of a Tripo character in Blender before rigging. Align the meshes and switch them discretely, shrinking inactive variants inside the head. Do not imply smooth blending or VRM compatibility.