Prompt Flicks

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.

Recreate a mobile-friendly browser game titled “Prompt Flicks,” a playable prompt-to-video toy built as a full-screen Three.js 3D scene with a fixed HTML HUD layered over it. The player types a natural-language prompt and presses “Create Video” (or Enter); there is no real online AI call. Instead, simulate local procedural generation with a short staged progress sequence, then rebuild the looping 3D scene by parsing keywords.

Visual presentation: use a dark, cinematic neon aesthetic with a nearly black/navy background, atmospheric fog, soft studio lighting, glowing cyan/magenta/yellow accents, and many small floating colored particles. The camera is a perspective camera positioned above and in front of the stage, looking toward the generated scene around y=2.2 and z=-3.1. Slowly sway the camera horizontally and vertically while a circular neon portal ring rotates and subtly pulses near the back. Place a large, low circular cylinder as the stage ground. Render with antialiasing, high-DPI support, ACES-like tone mapping, and responsive resizing. On the initial scene, show a floating generated title card near the upper center reading “cinematic space dragon video with glowing stars,” with a dark translucent rounded rectangle and cyan-to-pink-to-yellow outline.

HUD: at the upper left show the bold brand “Prompt Flicks,” with “Flicks” cyan. At upper right show rounded translucent pills for “Score 0,” “Clips 0,” and “Streak x1” (values update during play). Near the center/top show an initial challenge/help card: “Challenge: make a space dragon video” and explanatory text that the player can type any prompt and create a looping 3D video scene; hide this card after the first generated clip. At the bottom, show a translucent blurred rounded prompt dock with a text input and a bright gradient button labeled “Create Video.” The input placeholder is “Try: cinematic sunset city chase with robots.” On narrow/mobile screens stack the input and button vertically. Also show an AI-generation panel above the dock: model label “PF-DreamEngine v3,” a changing stage label, deterministic “Seed ####,” a gradient progress bar, and token chips. During generation, make the panel glow/pulse and change the button label to “Generating…” while disabling it.

Generation flow: on submit, use the typed trimmed prompt, or choose one of several built-in sample prompts if empty. Tokenize lower-case alphanumeric words, deduplicate, and display up to eight token chips, with the first chip prefixed by #. Compute a deterministic numeric seed from the prompt. Progress through these visible stages with short delays: “Tokenizing prompt” (12%), “Building latent space” (32%), “Diffusion pass 12/32” (52%), “Upscaling frames” (74%), and “Compositing video loop” (92%). Then render the scene, set progress to 100%, display “Clip rendered · AI pass complete,” and calculate a frame count based on token count. Show a brief rendering caption such as “Rendering: [prompt]”.

Prompt parsing and scene generation: choose the first matching environment style from these keyword groups: space (space, star, galaxy, planet, moon, rocket), ocean (ocean, sea, water, wave, fish, storm), forest (forest, tree, jungle, nature), city (city, street, car, chase, cyber, neon), desert (desert, sand, cactus), snow (snow, ice, winter), or lava (lava, volcano, fire); if no environment keyword matches, default to space. Apply a style palette to the background, fog, ground, portal, key light, and particles. Always add a prompt title card. Then add conditional procedural objects for semantic keywords: “dragon” creates a capsule-like dragon body, cone head, and flapping triangular wings; “robot” creates a small boxy robot head/body with glowing eyes; “alien” creates a large oval head, capsule body, and dark eyes; “car” or “chase” creates three glowing moving cars; “storm” or “lightning” creates five flashing lightning bars; “dance” creates eight orbiting colorful spheres; “cinematic” or “movie” creates two parallel glowing camera rails.

Environment details should match the actual procedural content: space has dozens of stars and four floating planets; ocean has animated torus-like wave segments and rising bubbles; forest has rows of conical trees and glowing mushrooms; city has neon-emissive box towers with window planes and a road; desert has flattened dune meshes and cactus trunks/arms; snow has falling small snowflake/icosahedron particles and ice crystal cones; lava has glowing crack bars, a cone volcano, and a pulsing volcano glow. Add a generic AI visualization layer to every generated clip: a moving horizontal scanner plane, a slowly rotating translucent wireframe latent-space cube, floating neural nodes positioned around it, and five blinking translucent frame-thumbnail planes near the floor.

Animate the generated clip continuously: planets float, stars twinkle, waves undulate, bubbles rise and reset, trees sway, mushrooms pulse, city emissive intensity pulses, snow falls and resets, dragon body bobs and wings flap, robots and aliens bob, aliens rotate/dance, cars travel across the stage and wrap around, lightning flashes, dance orbs orbit, dunes shimmer, the volcano rumbles, scanner line sweeps vertically, latent cube rotates, neural nodes pulse and drift, and frame thumbnails blink. Keep the generated objects clickable: raycast pointer/touch clicks against the generated 3D objects, store the selected object name, and briefly show a readable caption with the “generated” prefix removed and underscores converted to spaces.

Scoring and challenge loop: maintain score, clip count, streak, current challenge index, matched-last state, current prompt/style, playback time, and AI state. Use five rotating challenges: space dragon, ocean storm, sunset city, forest robot, and snowy alien dance. For each submitted prompt, score a base based on prompt length and unique word count, capped as in the original feel; award a large streak-scaled bonus when both challenge keywords are present. Increment clips; on a successful challenge increase the streak and advance to the next challenge, otherwise reset streak to 1. Update the challenge text and HUD pills after each clip. The scene loops indefinitely after rendering. Expose optional debugging helpers window.getGameState() and window.setGameState(patch) that report or adjust score, clips, streak, prompt, playing state, AI data, style, challenge, selected object, video time, and generated-object count.