BrawlBox – Turn a photo into a playable browser fighting-game character

Name: BrawlBox – Turn a photo into a playable browser fighting-game character
Availability: InStock
Author: tqliu_boston

by tqliu_boston·Jun 10, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerWizardryZero to OneCrowd Pleaser

Client-side photo-to-fighter pipeline with green-gap detection and atlas baking.

Strengths

•Client-side generation pipeline with no backend—your API keys never touch their servers
•Deterministic engine with fixed timestep enables reproducible matches and replays
•Full fighting game mechanics: throws, parries, charge motions, OTG, five CPU difficulties

Weaknesses

•Requires users to bring their own OpenAI and fal API keys, adding cost friction
•Fighting game audience is niche despite the impressive technical execution

Post Description

I built BrawlBox, a 2D fighting game that runs entirely in the browser, plus an AI creator that turns a single photo into a playable fighter — sprite sheet, animations, and a full 36-action moveset wired into the engine.

The part I'm most proud of is the pipeline: you bring your own OpenAI + fal keys, and the whole photo→fighter generation runs client-side — no account, no backend, nothing of yours hits my servers. Under the hood it re-skins a CC0 mannequin template, slices and keys the frames via green-gap detection, despills, scales, and bakes an atlas the engine can load.

The engine itself is deterministic (pure tick function, fixed timestep) so matches are reproducible — it started as a weekend project and got way out of scope (throws, guard/parry, charge motions, power meter, dizzy/knockdown/OTG, a CPU opponent with 5 difficulty levels).

Play it here: https://brawlbox.gg — happy to answer anything about the determinism model or the in-browser generation pipeline.