Building a screen capture library for Compose
summary
A screen capture library converts UI trees into compact wireframes rather than pixel screenshots, cutting a 100+ KB JPEG to 350 bytes payload (example screens ~430 bytes, ~60 elements) and reducing capture cost from ~50 ms and 10 MB memory to about 12 ms capture on traversal; works for Views and Compose by using accessibility/semantics APIs instead of private Compose internals, ships in an SDK with a control plane for dynamic enables, reaches tens of millions of devices and keeps P90 latency near the 16 ms frame budget.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:46Logging in production controversy
- 5:57Screen captures for mobile observability
- 8:28Traditional screenshot challenges
- 14:28Wireframe-based screen capture
- 17:22Real-time wireframe capture benefits
- 19:30Bit Drift logging system
- 22:56Extending to Compose views
- 27:35Semantic APIs for Compose capture
- 29:00Takeaways and future
- 30:28Live demo and Q&A
related talks
42mMaking UI Specs Visible — Compose Screenshot Testing and a Gallery to Support the AI Agent Era
35mEasy screenshot testing with Compose
34mHigh-Performance, Expressive Remote UI: Deep Di…
40mUnderstand the mechanism! Screenshot-testing Compose previews in many variations
54mOptimizing Render Performance of Jetpack Compose
24mDiscover and Reuse UI Components in Compose with Showcase
36mOptimizing Render Performance of Jetpack Compose
39mRe-Compose your UI: building an atomic design library