PixiJS 3D 🧊, UE Workflows 🎮, Browser Playtesting 🧪
Engines and Asset Workflows 🎮
PixiJS 3D Enters Open Beta
PixiJS 3D is now in open beta as an MIT-licensed 3D engine that integrates with PixiJS 2D on a shared stage. Developers can add 3D scenes as View3D display objects while reusing PixiJS workflows, assets and pointer events. The engine is designed for WebGPU and also runs on WebGL2, with built-in rendering features including animation, shadows and post-processing. It requires PixiJS 8.23.0 or later; its API may change before a planned 1.0 release.
Meshy workflow builds a cinematic environment in Unreal Engine 5
An Epic AI MegaJam demonstration uses Meshy to generate textured 3D assets from text prompts and reference images, then imports them into Unreal Engine 5 with materials configured through Meshy’s bridge plugin. The presenter assembles a space-station hangar using manual scaling, modular Blueprints and object scattering, and generates spacecraft bodies and thrusters separately for animation. Meshy also provides remeshing and polygon-count controls. The workflow is presented as a personal cinematic experiment; its suitability for games remains unverified.
Tripo demonstrates Agentic Canvas workflows for 3D game assets
Tripo demonstrated how its Agentic Canvas coordinates 3D asset generation, edits to individual parts, UV preparation, texturing and rigging through reusable node workflows. Examples included a treasure chest finished using Blender, Substance Painter and Unreal Engine, and a lunar mech shooter built in Unreal Engine 5 with a playable Windows build. The team emphasized that generated assets still require inspection and cleanup, and that facial details may need manual adjustment.
Fal.ai demonstrates AI-assisted game development workflows
Fal.ai product engineer Apek demonstrated an AI-assisted game development workflow spanning Three.js, Blender and Unreal Engine, using reference images, parallel level blockouts and automated scene dressing. Examples included the released browser games Viper 11 and Infinite Backrooms, plus an unfinished multiplayer FPS prototype. His downloadable Any Scene repository lets agents build Blender scenes from visual references. He emphasized that outputs still require human art direction, bug fixing and repeated iteration, and that refining image references before generating 3D assets can help control costs.
Builds and Playtesting 🛠️
BiteMe Games outlines its browser playtesting process
BiteMe Games argues for offering itch.io browser playtests when games are announced, then using feedback to refine them before releasing demos. Its D-Cube and Silicon Architect playtests prompted changes to controls, accessibility, pacing and balance. D-Cube recorded about 6,800 browser plays; its optimized build initially downloads roughly 30 MB and loads remaining assets in the background. The studio recommends frequent playtest updates and cautions that first-time developers may not achieve comparable results.
AWS and NetApp demonstrate FlexClone workflow for Unreal Engine builds
AWS and NetApp demonstrated an Unreal Engine build workflow using FlexClone on Amazon FSx for NetApp ONTAP to prepare agents without full workspace copies. In their cold-sync tests, agent preparation fell from 7 minutes 48 seconds to 8–11 seconds, and a build completed in 6 minutes 28 seconds—3.2 times faster than stock Horde, or about twice as fast as an optimized setup. They noted that conventional syncing can cost less for low-volume farms; the public Cloud Game Development Toolkit examples are starting points, not production designs.
AI, Bugs, and Game Design 💬
Hopfield and BAM Networks Can Be Trained and Run with Integer Arithmetic
A tutorial by Alan Wolfe shows how to train and run Hopfield and Binary Associative Memory (BAM) networks by hand using binary patterns converted to bipolar values and integer matrix operations. Hopfield networks recall stored patterns from noisy inputs, while BAM networks associate input-output pairs in both directions. Worked examples demonstrate recall, but the article also notes that Hopfield inference can enter loops and that storing more patterns can reduce recall reliability.
Wookash and guests play Big Walk and discuss AI use in games
Wookash, Anton, Valigo and Raphael Luba play the multiplayer game Big Walk, whose proximity-based voice chat and gestures shape how players coordinate while exploring and solving puzzles. The stream also discusses controversy over the game’s AI disclosure: participants say developers used Claude to address an obscure Unity bug, disputing claims that the game was made entirely with AI. They debate AI’s role in game development and offer mixed views on Big Walk’s puzzles, controls and value as a social game.