This update adds the most requested thing from the discussions: more than one translation region at a time. Nothing changes if you keep using a single region or the whole screen.
NEW: Up to 4 translation regions. Press Shift+F5 to add another region while keeping the ones you already have (F5 still replaces everything with a single region, as before). Each frame has its own close button and can be dragged on its own. F6 translates all regions at once, and auto mode watches all of them, but only the regions where something actually changed are read. Typical use: the dialogue box plus a quest log or chat panel in a corner.
HOW TO: Select the first region with F5 as usual, then press Shift+F5 and drag the next one. The toolbar shows "Region 2/4" and so on. The close button on a frame removes only that frame; the Clear key with no translation on screen removes all of them. The add key can be changed in Settings under Hotkeys. If you prefer no frames at all, turn off "Show selected region" in Settings > Display; regions keep working without the frames.
IMPROVED: Multiple regions are read in a single recognition pass, so they cost less than a full-screen pass. Measured on real game footage: recognition time per job was 346 ms for the full screen, 208 ms for one region, 221 ms for two regions and 271 ms for four, with no increase in missed text. The measurement harness is in the repository.
NOTE: Regions are remembered for the current session only, as before. If anything looks wrong with multiple regions, the multi_roi switch in config.json (display section) restores the single-region behavior. The Intel NPU/GPU (OpenVINO) build remains on the "beta" branch.
