KineForge

Cinevva's analyzer read this game's design doc and its source, then wrote the brief below. "Make your own" hands it to the game creator as your opening message, so you start from a real spec rather than a one-line idea.

Rebuild “KineForge,” a browser-based interactive 3D scene builder/editor rather than a conventional score-based game. Use a dark navy/black interface with neon violet and teal accents, translucent rounded panels, thin blue-violet borders, and a clean Inter-style sans-serif UI. The app fills the screen with a top toolbar, a central Three.js-style 3D viewport, a left Outliner, a right Properties inspector, and a bottom AI prompt dock.

In the viewport, render a perspective 3D scene with a dark blue ground plane, visible square grid, soft fog, hemisphere ambient lighting, a strong directional key light, and real-time shadows. Start with two editable assemblies already placed on the floor: a low-poly wooden chair on the left and a snowman on the right, matching the reference view. The default camera is an orbiting perspective camera positioned above and to the side, looking toward the objects at roughly a 45-degree view. Users can drag to orbit, scroll or pinch to zoom, and click/tap visible meshes to select their parent assembly. Show a small pill hint in the viewport reading “Drag to orbit • Scroll to zoom • Tap objects to select” and a “perspective · 45°” label in the lower-right.

The top bar should show a K logo, the title “KineForge,” and subtitle “Scene Builder.” Provide primitive creation buttons for Box, Sphere, Cylinder, and Cone. Each adds an editable colored primitive to the scene: boxes are violet, spheres teal, cones gold, and cylinders violet-toned, positioned in an automatic spread. Add transform mode buttons Move, Rotate, and Scale; the selected assembly displays a transform gizmo and the modes can also be switched with W, E, and R keyboard shortcuts. Include Duplicate, Add Keyframe, Play/Pause, and red-accented Clear controls. Delete or Backspace removes the selected object; Ctrl/Cmd+D duplicates it with a small x offset.

The left Outliner lists every created top-level assembly with a live object count badge. Each row has the object name and index, highlights the currently selected object in teal, and shows an animation indicator when applicable. If empty, display “Empty scene. Add a primitive or ask the AI.” Selecting either a viewport object or outliner row updates the active object and inspector.

The right Properties panel edits the selected object. Include an editable name field, numeric Position controls for X/Y/Z, Rotation controls in degrees for X/Y/Z, and Scale controls for X/Y/Z. Include Focus and Delete buttons. Add an Animation card showing the current animation state and preset buttons: Spin, Bounce, Float, Pulse, Orbit, and Launch, plus Add Keyframe and Clear Anim. Transform edits should apply directly to the selected group. Focus reframes the orbit camera around the selected object.

Implement looping animation presets on selected assemblies. Spin rotates around Y; Bounce repeatedly raises the object and returns it; Float gently bobs vertically; Pulse expands and contracts its scale; Orbit moves it around a horizontal circle; Launch makes it rise substantially while spinning. Presets can be combined, and Play/Pause controls all active animations. Add Keyframe records the selected object’s current position, rotation, and scale, pauses playback, and lets the user move it before recording another pose; interpolate smoothly between recorded keyframes when played. Show animation labels such as “Spin + Pulse” or “2 keyframes.”

At the bottom, add an AI dock with a pulsing teal status dot, “KineForge AI ready,” and a message explaining that the local AI builds editable 3D parts. Provide suggestion chips for: “Build me a small wooden chair,” “Create a rocket launch animation,” “Bounce selected object,” and “Spin and pulse selected.” Add a large prompt input with placeholder text about making the selected object spin, bounce, or fly upward, plus a Forge button. The prompt interpreter is local, deterministic, and does not use an external backend. After a short thinking delay, parse prompt keywords to create assemblies and/or animate newly created or currently selected objects.

Support these prompt-built assemblies from primitive meshes with simple stylized low-poly construction and appropriate colors: wooden chair (seat, back, four legs and dark wood details), snowman (three white snow spheres, black top hat, orange carrot nose, coal eyes, brown stick arms), low-poly rocket (metal cylindrical body, red cone nose and fins, blue window, orange flame), humanoid (skin head and arms, blue torso, dark pants, coal-colored eyes), treasure chest (dark wood box, curved wood lid, gold bands/lock and small gold details), pine tree (brown trunk and two green cone layers), tiny cabin/house (wooden body, dark red roof, door and blue windows), campfire (stone ring, crossed logs, glowing flame and orange point light), mini castle (stone body and towers with dark red roofs, door and windows), and tiny car (red body, blue cabin, four dark wheels). If a prompt contains “forest,” create six pine trees; “many” creates four matching objects; “two” creates two. Recognize related words such as seat/stool, snow, spaceship, person/character, loot, hut/home, fire/bonfire, tower/fort, and vehicle/truck. Unknown non-animation prompts should create a colorful abstract sculpture made from up to seven boxes, spheres, and cones arranged vertically/diagonally and named from the first prompt word.

Recognize animation language in prompts: spin/rotate/turn/twirl, bounce/jump/hop, float/hover/bob, pulse/grow/breathe, shake/wiggle/wobble, orbit/circle/around, launch/fly/rocket/blast/upward/take off, and dance (combine bounce, spin, and pulse). Phrases such as “animate,” “move,” or “keyframe” should also trigger animation behavior. If a prompt creates objects and requests animation, animate all created targets; otherwise animate the currently selected object. If animation is requested with no target, tell the user to select or create one. Frame the camera automatically around newly built objects. Keep the scene editable throughout, with no score, enemies, combat, lives, or win condition; the core goal is assembling, transforming, inspecting, and animating simple 3D scenes through direct manipulation and deterministic natural-language commands.