v0.4.0

Banana Slides v0.4.0 brings internationalization, dark mode, a redesigned editor experience, and dozens of quality-of-life improvements.

---

Highlights

✏️ New Editor Experience

• MarkdownTextarea — A brand-new rich editing component with inline chips, formatting toolbar, image preview, and markdown-to-chips conversion on paste. Now user can paste images whereever in homepage, outline editor or description cards and they will be processed instantly.
• OutlineEditor Redesign — Split-view layout: left panel for editing raw outline/description text, right collapsible drawer for structured outline cards.
• Glassmorphism UI — Modal and MaterialGeneratorModal redesigned with a modern glassmorphism aesthetic.

🧮 LaTeX & Image Paste

• Inline ($...$) and block ($$...$$) LaTeX formula rendering via KaTeX in all Markdown views.
• Paste images directly from clipboard in description editors — they are auto-uploaded as materials and inserted as markdown image links.
• Uploaded images now get AI-generated captions (Gemini / OpenAI) used as alt text.

🖼️ Material Center

A dedicated Material Center modal for browsing, filtering by project, batch selecting, bulk downloading (as zip), uploading, deleting, and previewing materials.

⚡ Performance

Frontend image loading optimized with compressed JPG thumbnails (cached_image_path). Originals are preserved for export quality.

🌍 Internationalization & Dark Mode

Full Chinese/English bilingual support powered by i18next, with light / dark / system theme switching. Every component has been adapted for dark mode.

---

Features

• i18n + Dark Mode (#196)
• MarkdownTextarea, OutlineEditor redesign, glassmorphism UI (#206)
• LaTeX rendering, image paste, and AI captions (#203)
• Material Center and Help Modal — Paginated onboarding guide, auto-shown on first visit (#53)
• Image cache optimization — Thumbnail generation for faster frontend loading (#151)
• Export quality assurance — "Allow partial export" toggle (default off), fail-fast with detailed error messages and help text (#190)
• Settings service testing — One-click connectivity tests for Baidu OCR, text model, caption model, image generation, and MinerU PDF parsing; supports testing unsaved settings before saving
• Editable outline & description in SlidePreview — Edit title, points, and description directly in the preview modal
• Export outline & description to Markdown
• Manual chapter (part) editing in outline (#194)
• AI reasoning mode — Independent enable_reasoning toggles for text and image generation; Vertex thinking_budget support (#160)
• Resolution mismatch detection — Warning dialog when image generation resolution doesn't match settings; OpenAI extra_body params for resolution control
• Docker multi-arch — amd64 + arm64 builds
• Pre-built Docker image deployment option
• PPTX metadata — Author and date metadata in exported files

Bug Fixes

• Export directory not created — Relative path caused PDF/PPTX export to fail with 500 (#201)
• Reference files not available during generation — Files were associated after AI generation completed; now associated before (#201)
• Image generation page data mismatch — Filtered page_ids caused outline data to be zipped with wrong pages (#201)
• Stale closure in task polling — Switching projects caused polls to query the wrong project (#201)
• Infinite polling on description generation failure — Added retry limit (#201)
• Path traversal vulnerability — Export filenames and /files/ paths sanitized with secure_filename / os.path.abspath (#201, #206)
• Export path double-join — get_absolute_path() wrapped an already-absolute path, causing 500
• Memory leak — Object URLs not revoked in image previews (#201)
• Description generation race condition — Now uses API response data directly instead of reading from store
• Database session staleness — Expire session before export to prevent stale reads (#183)
• Baidu OCR config — Read API key from Flask config (database settings) instead of only environment variables (#190)
• Silent error swallowing — MinerU / Baidu OCR extraction failures now propagate to frontend (#190)
• Image path sync — Added retry mechanism in pollImageTask
• Batch delete — Use Promise.allSettled to handle partial failures (#201)

Improvements

• Hover tooltip with format examples for outline/description creation tabs (#202)
• HTTP error codes (503, 500, 429, etc.) show user-friendly localized messages (#202)
• Modal: always show close button + Escape key support (#202)
• Toast: new warning type, improved stacking, no header overlap (#202)
• Stale project IDs silently cleaned from localStorage (#202)
• Removed intrusive "service test tip" toast on first project creation (#202)
• PORT renamed to BACKEND_PORT for consistency
• Default text model updated to gemini-3-flash-preview

Infrastructure

• GitHub Actions workflow for automatic README Chinese → English translation
• E2E tests adapted for i18n (locale=zh-CN)
• Test fixtures updated for open-source codebase (#204)
• Docker: IN_DOCKER=1 set in backend Dockerfile

New Contributors

• @VVIP1024 made their first contribution in #160
• @uthanw made their first contribution in #151
• @Rock-ql made their first contribution in #53

Full Changelog: v0.3.0...v0.4.0