Fix(UX/AI): implement popstate listener to handle Android hardware back button closing modals instead of the app, and replace AI-generated inline image URLs with real Tavily web images via placeholders to completely eliminate Markdown parser errors
Feat(AI): switch primary image source to Tavily for authentic real-world images instead of AI generated ones, falling back to Wikipedia and Pollinations
Fix(AI): simplify Gemini prompt by enforcing hyphens instead of URL encoding for inline images and restricting Mermaid label syntax to prevent Markdown parser breakages
Feat(AI): explicitly instruct Gemini to generate 1-2 inline images per article using strict Pollinations URL format with random seed to ensure uniqueness and visual appeal
Fix(UI/Backend): completely remove delete confirmation logic for instant optimistic UI updates and replace static Wikipedia image fallback with dynamic AI generated image to fix repeating images
Feat(AI): replace slow pollinations AI image generation with ultra-fast real Wikipedia cover image fetching, and remove unreliable inline image prompts
Fix(AI): add random seed to pollinations image generation to prevent repetitive generic faces and properly encode markdown inline image URLs to prevent broken images
Feat(UI): add 'generate news' button permanently to the top of the news feed so users don't have to wait for empty state, and add an article delete button inside the article view to remove unwanted articles
Fix(UI): correctly assign the generated AI image to the article object in news-service and add custom marked renderer combined with mermaid.js injection to properly render mermaid graphs inside news articles
Feat(AI): prompt Gemini to include dynamic AI generated images using pollinations.ai in both the news article covers and directly embedded inside the news article content alongside markdown diagrams
Fix: Change Gemini model cascade to use official gemini-2.5-flash and gemini-1.5-flash to prevent hallucinated endpoints returning truncated garbage, and remove maxOutputTokens limit