1.0.0

Full Changelog: v0.9.0...v1.0.0

0.9.0

What's Changed

This release rolls up everything since v0.6.0 (including the unreleased 0.7.0 and 0.8.0 work).

New Features

• Nano Banana 2 model — icon generation now supports Google's banana-2 model (#28)
• GPT Image 2 model — added OpenAI's gpt-image-2 to the available model options (#32)
• Feature graphic command — new feature-graphic command for generating Google Play Store banners (#30)
• Custom OpenAI base URL — configure openai_base_url via config or CLI flag to use OpenAI-compatible endpoints (#35)

Fixes

• Fixed CLI error double-wrapping, a config read side effect, and stale flag docs (#34)
• Fixed common issues in the feature-graphic flow

Full Changelog: v0.6.0...v0.9.0

0.6.0

[0.6.0] - 2026-02-11

🤖 Multi-Provider Support & Prompt Preview

New Features

• Google Gemini Integration: Generate icons with Gemini alongside OpenAI

  • --model banana for Gemini 2.5 Flash Image generation
  • --pro flag to use Gemini 3 Pro Image Preview for higher quality
  • --google-api-key flag and GEMINI_API_KEY / SNAPAI_GOOGLE_API_KEY env vars
  • Quality tiers for Gemini Pro: --quality 1k|2k|4k

• Prompt-Only Mode: Preview prompts before spending API credits

  • --prompt-only flag shows the final enhanced prompt/config without generating images
  • Works with all flags (--style, --raw-prompt, etc.) for easy prompt iteration

• Google API Key Management: snapai config --google-api-key to store and manage Gemini keys alongside OpenAI keys

• GitHub Actions CI/CD: Automated npm publishing workflow

  • Publishes on GitHub release events
  • Automatic beta dist-tag for pre-releases
  • Manual workflow_dispatch trigger with custom dist-tag

Improvements

• Updated Model Naming: Clearer model aliases

  • gpt-1.5 (default, best quality), gpt-1 (fast), banana (Gemini)
  • Legacy alias gpt still supported for backward compatibility

• Refactored Prompt Engine: Shared context and technical rules for more consistent icon output

  • Design constraints prevent common failures (no rounded corners, no UI elements)
  • Better platform-specific prompt enhancement
  • Style applied as dominant constraint when combined with --raw-prompt

• Expanded Style System: Style flag now accepts any string hint, beyond the built-in presets

  • Massively expanded style templates with richer descriptions
  • Dangerous style detection blocks photorealistic/portrait requests

• Improved CLI Flags:

  • -r shorthand for --raw-prompt
  • -s shorthand for --style
  • -i / --use-icon-words to optionally include "icon"/"logo" in the enhancer
  • -n shorthand for image count (replaces --num-images)
  • -k / --openai-api-key replaces --api-key (old flag kept as hidden alias)

• ESLint Configuration: Added .eslintrc.cjs for consistent code quality

Breaking Changes

• Deprecated size options removed — output is always 1024x1024
• --api-key flag renamed to --openai-api-key (old flag still works but is hidden)
• --num-images flag replaced by -n / --n (old flag still works but is hidden)
• Icon composer command and related utilities removed

Technical

• New GeminiService with streaming support and binary image extraction
• buildFinalIconPrompt utility for unified prompt generation across providers
• Google API key validation in ValidationService
• TypeScript interfaces for Gemini options and binary image types
• Removed pngjs dependency