Pixel-art baseball game comparison
- Views
- 7.8K
- Likes
- 163
- Replies
- 21
- Saves
- 96
A single-file pixel-art 3D baseball management game on a floating voxel island, built separately by Opus 5.5 and Sonnet 5.5 from the same prompt. The creator compares time and cost for each model.
Full prompt6,645 chars
Build a single-file HTML5 game, "Marigold Park": a pixel-art 3D baseball game on a floating voxel island, where you manage one of two teams (you choose the approach, the pitches, the defence and the steals; the sim plays out each pitch). three.js r128 from cdnjs, no build step for the player, everything inline, no asset files. LOOK Render the 3D scene to a low-res target (default 2 px art pixels, slider 1-6), then a post pass: depth-based outlines, snap every pixel to a fixed 90-colour OKLab palette, Bayer dither. Orthographic isometric camera (rotate in 45 degree steps or drag, zoom, tilt) placed 850 m from its target with a far plane of 1700, so the island is never clipped at low tilt. Sun shadow map, day / dusk / night with eased light changes and floodlights that come on after dusk. Island: a slab of 2.4 m voxel columns (about 208 x 192 m), terraced rim, banded soil and slate, a waterfall. The slab is solid right through the centre (every cell included) with DoubleSide faces, so no hole shows from any camera angle. A 2048 px painted ground: outer grass patchwork, the park (mown stripes, fair and foul ground, warning track, infield dirt skin, grass diamond, base paths, circles, chalk lines), and the village (paths, plaza, parking, playground, pond, stream). Park geometry in metres, home plate at the origin: bases 18 m apart, mound 13.5 m out, fence 60 m at the lines swelling to 70 m at centre (3 m high), foul poles, backstop, two dugouts with benches, bullpens, umpires, base coaches and an on-deck batter. Three playing surfaces that repaint the ground and change the physics: natural grass, artificial turf (the ball skips and keeps rolling, fewer bad hops), red clay infield (slow and heavy but more bad hops and errors). Stands in eight side groups around the park: coloured seat sections, a brick concourse wall with pilasters, windows and a cornice, a striped canopy with a scalloped valance over the back rows, a press box, an outfield scoreboard, outfield bleachers beyond the fence, advertising boards, banners, bunting, rails, pennant poles at the corners, vendors in the aisles, a mascot by the third-base dugout, floodlight towers, thousands of instanced spectators in team colours who cheer on big plays. Village around the park: plaza with a gate arch and sign, ticket booths, a statue of a batter on a pedestal, a fountain, market stalls with umbrellas, benches, lamps, hedges, planters, a car park, a playground, a pond, a stream with a waterfall, groves, walkers and floating islets. Doll's-house lift: the eight stand sides that face the camera rise out of the way (with their crowds and fence) and ease back when the camera turns; disabled for cinematic cameras and between innings; toggle "Lift near stands". GAME Six invented teams (Harbor Gulls, Cedar Bears, Ironwood Foxes, Riverton Rays, Maple Hawks, Dunes Lizards) with generated rosters: 9 hitters (C, 1B, 2B, SS, 3B, LF, CF, RF, DH) with contact, power, eye, speed, range, arm, glove and handedness (R / L / switch), plus 4 pitchers (a starter with 70-90 pitches of stamina and three relievers with about 30) with velocity, control, movement and a repertoire from fastball, slider, curve and changeup. Block-built people with caps, helmets, gloves, bats, catcher's gear, umpires; poses for windup, delivery, swing, bunt, run, slide, throw and catch. Formats: Quick (3 innings), Short (5), Full (9). Ties after the last inning go to extra innings (the previous batter starts on second base, up to three extra innings), walk-offs end the game on the spot. Fixed-step 60 Hz sim. Each pitch: windup, release, flight with break per pitch type (fatigue worsens control past about 80% of the pitcher's stamina), a strike zone 0.6 m wide and 0.78 m tall, umpire calls with a small error rate on borderline pitches. Challenges: each side has 2; when a borderline call goes against the team you manage you get a 4.5 s window to challenge (key C); an electronic strike-zone overlay then shows the truth; the AI challenges wrong calls with high probability. Batting: approach Contact / Power / Patient / Bunt, aim Auto / Pull / Center / Oppo. By default the sim swings for you using timing windows scaled by the hitter's contact skill; or tick "Time the swing myself" and press Space with windows widened. Batted-ball physics with gravity, drag, wind, bounces, rolling, fence hits and home runs; wind modes Calm, Out to centre (+8% carry), In from centre (-10%); the surface changes bounce and roll. Fielding and running: fielders take positions by defence setting (Normal, Infield in, Deep); the sim solves interceptions with range, arm and glove, picks the target base, throws along arcs, resolves force and tag plays, double plays, errors and bad hops, tag-ups and sacrifice flies, foul balls, home runs robbed at the wall, steals (key T) and wild pitches, walks, strikeouts. Runners follow real base paths with leads and slides. Pitching side: choose pitch type (Mix, Fast, Slide, Curve, Change), target (Mix, Zone, Corners, Chase), defence, and change pitchers from a drop-down. The AI manager handles the side you do not manage. Flow: ceremony and line-ups, innings with an inning-break caption, play descriptions after every ball in play, final result with the walk-off note. Auto-next game option. UI On the canvas: a broadcast bug (inning, outs, count, runners on a diamond, scores, current batter and pitcher with pitch count), pitch speed and type read-out, captions for big plays, a challenge button with a countdown and the "eye" overlay, a swing button for touch. Panel: game length, team pickers, which team you manage, new game, skip ceremony, speed (1/2x, 1x, 2x, 4x), auto-next, pause; surface, wind, time of day, floodlight state; coach controls for batting or fielding; camera (TV follow, Behind plate, Island), rotate, zoom, pixel size, outline / dither / raw / sound / lift toggles; box score for both teams, play-by-play feed. Keys: Space swing (manual) or pause, N new game, K skip, P pause, C challenge, T steal, S surface, W wind, L light, Q/E rotate, +/- zoom, R reset the view, 1-3 cameras. Phone: coach controls and swing button docked under the canvas, no horizontal scroll. Procedural WebAudio (bat crack, mitt pop, crowd roar, groans). QUALITY BAR Headless bot test: full 9-inning games finish with plausible lines (roughly 3-10 runs a side, a few home runs, some errors, a few steals), no stuck states, no console errors, no horizontal scroll on a phone, no holes in the island from any camera angle. No personal data, keys or paths in the source; the only external requests are the three.js script and Google Fonts.
Source: the creator's reply under the post. Kept in its original language, unedited.
Want more prompts like this one?
Get the 50 most-viewed Claude 5.5 video prompts on one page: the full original text, the creator and the post it came from. We'll only email you when there's a new prompt worth trying. Unsubscribe anytime.
Tools the creator mentioned
Original post: https://x.com/vib3coded/status/2118281844181369210
All works are open data at GitHub · awesome-opus-5.5-video.
Model attribution is as stated by the creator and was not independently reproduced. View counts are a snapshot from 2026-11-08. The work belongs to its creator; DM @GoSailGlobal on X for corrections or removal.
More comparisons
0:42AI self-portrait challenge: Sonnet 5.5 vs GPT-6 Astra
7.3M views
0:28Opus 5.5 vs GPT-6 Astra samurai game demo
1.1M views
0:13Gummy bear soft-body physics scene
631.5K views
0:30GPT-6 Astra vs Claude Opus 5.5 comparison
605K views
0:22Grok vs Opus grass-touching simulator cost test
554.9K views
0:44Opus 5.5 vs GPT-6 Sol Unreal game dev
544.2K views
Related articles
OpenAI Academy vs Anthropic: A Complete Comparison of Two Leading AI Companies' Free Learning Paths
Both of the top AI companies offer official free learning platforms — together that's 25+ courses and hundreds of hours of content, enough to take you from AI newcomer to resume-worthy engineer. This post maps out all 15 OpenAI Academy courses along a 10-week roadmap and provides a full side-by-side comparison with the Anthropic learning path.
My Current AI Tools List | 2025 Edition
The AI tools I shared at a corporate training session — if you can't be bothered sorting through options, just use the best ones.
The Complete Map of Free AI Learning: 10 Official Platforms, $0 Total Cost
Anthropic, OpenAI, Google, NVIDIA, Microsoft, AWS, Meta, IBM, DeepLearning.AI, Hugging Face — every official AI course from all 10 platforms is free. This is the master index: one comparison table + learning paths by role.
Indie Developer vs. Businessperson: A Developer Who Can't Market Is Just a Backend Contractor
Indie developers chase interesting problems to build their dream products, while businesspeople repackage those same products and sell them directly to customers. A developer who can't market is nothing more than a backend contractor.