Cheng Lou

Cheng Lou is a software engineer at Midjourney and the creator of React Motion and Pretext, open-source projects that reimagine how interfaces move and how text occupies a screen. His work across React, Messenger, ReasonML, ReScript and generative AI explores how software can become more expressive without diminishing human creativity or control.

Lou established his approach to interface engineering with React Motion, which uses spring-based interface animation to generate natural movement from physical properties such as stiffness and damping instead of manually specified animation curves. His 2016 examination of software abstraction articulated a complementary principle: general-purpose tools become counterproductive when their flexibility burdens developers with complexity irrelevant to the actual problem.

At Midjourney, that attention to usable, responsive interfaces informs a broader account of how AI changes creative work. His AI Engineer World’s Fair 2024 talk traced unexpected technological consequences from graphics processors repurposed for machine learning to game-playing systems that made expert instruction more accessible.

More recently, Lou created Pretext, a JavaScript and TypeScript library for DOM-free text measurement and layout. By preparing text segments and reusing cached widths, it reduces repeated browser layout work while supporting multiline text, changing container sizes and unconventional arrangements. In introducing the project, he described accurate, efficient text measurement as infrastructure for more ambitious interfaces.

• AI-assisted human learning: Chess and Go engines did not eliminate interest in those games; they expanded access to high-quality instruction. Lou applies that logic to drawing tools whose assistance can recede as users develop independent skill.
• Human-centered creative instruments: AI could translate expressive hand gestures into intricate musical detail or help interpret an unfinished drawing, creating new forms of direct manipulation while preserving experimentation and authorship.
• Context-aware interaction: A stylus stroke should be interpreted alongside nearby objects, recent actions and conversation, replacing rigid gesture rules and unnecessary tool switching with systems more responsive to human intent.
• Designer-curated adaptive interfaces: Generate and evaluate many interface layouts offline, let designers select acceptable options, then use lightweight classification to choose among them instantly according to user intent or experience.
• Emergence from simple components: Resource-intensive autonomous agents limit the number of participants a system can support. Large networks of simpler computational units can instead produce useful collective behavior unavailable to a handful of individually sophisticated agents.