Skip to content

Shotluma 0.2.0

Choose a tag to compare

View all tags
@realZachi realZachi released this 08 Aug 14:51
· 32 commits to main since this release
v0.2.0
724e6c9

Official website launch, major performance improvements, screen-spanning mockups, and enhanced AI efficiency

🌐 Now Live at shotluma.com

Shotluma now has an official home at shotluma.com with a dedicated landing page showcasing features, demos, and getting-started guides. The editor runs directly at shotluma.com/app — no account, no backend, 100% local-first in your browser.

🚀 Performance & Efficiency

Up to 2× faster AI generation and 50% lower token costs with the same output quality. The AI runner now intelligently batches tool calls and employs provider-specific prompt caching strategies:

  • Anthropic: Explicit cache breakpoints reduce input costs by ~90% on Claude runs
  • OpenAI: Per-run cache keys improve GPT-5.6 prefix matching and hit rates
  • Google Gemini: Optimized for implicit caching with ~50% lower input token usage

Element measurements are rounded to one decimal to reduce payload sizes across accumulated conversation history — a small change that compounds across every turn of the agentic loop.

The prompt now instructs models to batch independent operations (entire slide compositions, repair rounds, final preview passes) into single turns rather than making individual calls, cutting round-trip latency and context re-reads significantly.

✨ New Features

Screen-spanning device mockups

Device frames can now span across adjacent artboards for continuous multi-screen narratives. Enable the new Spans screens toggle in the element toolbar to let a mockup flow seamlessly from one screen to the next — perfect for showcasing app flows and onboarding sequences across your screenshot set.

The spanning behavior respects canvas boundaries and integrates with drag constraints, nudging, and alignment. Ghost overlays visualize the continuation on adjacent screens while editing.

🔧 Improvements

  • AI prompt refinements: The generation prompt now explicitly teaches batching patterns and marks rarely-needed tools as such, guiding models toward more efficient execution paths
  • Improved API key management: Clearer dialogs and better handling of browser-stored provider keys
  • Enhanced documentation: Architecture docs now detail the prompt caching layer and cost optimization strategies
  • Updated repository links: All references now point to the canonical github.com/realZachi/shotluma

📦 Under the Hood

  • New src/ai/prompt-caching.ts module encapsulates provider-specific caching logic with moving Anthropic breakpoints and OpenAI cache-key injection
  • src/editor/screen-span.ts implements the spanning geometry and neighbor detection
  • Expanded test coverage for drag bounds, nudging, and screen-span interactions (vitest coverage remains >94%)

🐛 Fixes

  • Hardened browser provider key handling to prevent leakage in edge cases
  • Improved mobile layout on the landing page

Full Changelog: v0.1.0...v0.2.0

Assets 2
Loading