v0.5

[volfiros]

Summary

Performance

• Concurrent panel processing: Replaced sequential translation with a fire-and-forget + streaming architecture. The content script sends all panel data at once, the background processes up to 5 panels concurrently, and streams results back to the tab as each panel finishes — overlays appear incrementally instead of all at once
• Batched bubble translation: All speech bubbles detected in a single panel are now sent to Gemini in one grouped request (chunked at 16 images) instead of sequentially per bubble, significantly reducing API calls and latency
• Full chapter translation: The translate button now processes all manga panels on the page instead of a single visible panel

Features

• Stop translation: Clicking stop halts the background batch mid-flight — already-translated panels remain visible on the page
• Rate limit abort: When a rate limit is hit during batch processing, the entire batch stops immediately and the user is notified
• Token + time logging: At the end of each batch, the background console logs total time elapsed and token breakdown — input, output, and combined total
• Type-aware bubble translation: text_free bubbles (sound effects / free-floating text) are labelled separately in the Gemini prompt and translated as onomatopoeia, while speech, narration, and tall types are translated as dialogue

Fixes

• Rate limit toast no longer shows "Translation complete" when the fallback model is used after the primary hits a rate limit
• Stop translation preserves existing overlays rather than clearing them

Version

• Bumped to 0.5.0

Test plan

• Open a webtoon with 10+ panels and click Translate — overlays should appear panel by panel as results stream in
• Verify background console shows token breakdown and elapsed time after batch completes
• Click Stop mid-translation — processing halts, already-translated panels stay visible
• Trigger a rate limit — batch stops immediately with a rate limit toast, not a success toast

[Copilot]

Pull request overview

This PR updates the extension to v0.5.0 and reworks translation into a batch + streaming architecture so whole-page (chapter) translation can run concurrently and render overlays incrementally, with stop/rate-limit handling and token/time logging.

Changes:

• Switch from single-panel sequential translation to full-page batch processing with concurrent background execution and per-panel result streaming back to the content script.
• Batch speech-bubble translation per panel by sending grouped bubble crops to Gemini (chunked) to reduce API calls and latency.
• Add rate-limit + stop behavior updates, improve error/toast messaging, and log token/time usage after batches.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
ui/src/types/translator.types.ts	Renames bubble type interface to DetectedBubble for detector/model pipeline.
ui/src/popup.tsx	Improves error toast title for rate limits; adjusts status handling and some button border styles.
ui/src/contents/translator.tsx	Replaces sequential per-panel processing with batch dispatch + streaming overlay application and batch completion handling.
ui/src/contents/components/TranslationOverlay.tsx	Simplifies overlay styling and updates minimum font size.
ui/src/background.ts	Adds concurrent batch processing, stop support, detector/model integration renames, and batch token/time logging.
ui/src/api/services/gemini.ts	Adds token accounting, bubble-crop batch translation, and rate-limit/fallback tracking flags.
ui/package.json	Bumps extension version to 0.5.0.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.