v1.0.0 v1.0.0: first stable – VR gallery, wrist button, background service, French

[v1.0.0] – 2026-10-02

First stable release. Update: AUR yay -Syu linuxvr-viewshot or the new AppImage – open the app once, it copies the new layer + background service to ~/.local by itself (or press ⬆ Update layer). From a clone: 🔧 Rebuild & install. Then restart the VR game.

• 👋 Welcome on first start: three short windows – the app is only for setup/installing, afterwards everything happens in VR · install now (🔧 Install now / Later, skipped if already installed) · set up your translators → OK opens Options → Translation. Shown once (welcome_done in ui.json)
• 🇫🇷 French as app language (Options → General → Language), all texts in UI/core/i18n_fr.py; missing texts fall back to English. The VR panel of the background service speaks French too
• 🔘 Button to open/close the panel (like WayVR) – drawn by the layer on the same anchor as the panel (default: on top of the anchor hand, 3 cm, right hand mirrored; head: bottom left; world: in front). Laser + trigger toggles the panel (white ring = open), state is kept in panel_state.json. Edit mode: button and panel move/rotate (grip) and resize (bottom right corner + trigger) separately – own button_pose.json; a new panel appears above the button. ⟲ resets both. Options → Shot and ⚙ in VR: button on/off, Open when translating (closed panel opens after a photo, on by default), button size (2–20 cm), button colour (7 colours in VR, colour picker in the app).
• Faster size switch between translation and gallery: the layer checks panel_pose.json every 80 ms instead of 500 ms
• 🖼 VR panel: pages + gallery – at the bottom ◀ left / ▶ right switch between 🌐 translation and 🖼 gallery (● ○ shows where you are), ⚙ stays at the top. Size per page: translation and gallery each keep their own width/height in VR (panel_sizes.json), the gallery is bigger by default (48 cm) and opens big right away. The gallery fills the whole panel: 4 thumbnails per row with month dividers, ▲ / ▼ pages next to ⚙. Tap = photo large with 📋 Copy · ☁ Upload + link · ↗ Share · 🌐 Translate · ⓘ Info · 🗑 Delete (tap twice → trash) and ‹ ›. Same in the app and in the background service (it copies via wl-copy/xclip, uploads to directupload.eu, shares, moves files to the trash itself). The app's panel works right away
• 📁 Gallery folders – ⚙ in the desktop gallery: photo folder (fixed) + more folders (e.g. ~/Pictures/VRChat), change / remove, include subfolders for each. Desktop + VR gallery show all of them, newest first; only ViewShot photos are translated. Thumbnails load in the background (no freeze with thousands of screenshots)
• 📅 Month dividers in the desktop and VR gallery (“──── 2026 October ────”)
• Options → Translation: Settings for dropdown – set up any service without switching Main; 🧪 Test checks the service shown
• Service dropdowns grouped: “── Translation ──” (Lingva, Google, LibreTranslate, DeepL, own API) and “── AI translation ──” (Claude, Gemini, ChatGPT, own command, image LLM) – main page, options and VR panel
• 🕘 History on the main page moved to Options → General (card Main page)
• Fix – share to Discord: Vesktop / Discord / WebCord are also found as Flatpak (e.g. dev.vencord.Vesktop)
• 🪟 Panel: ⚙ settings instead of the edit/reset buttons – they slide down from the top like on a phone (a few frames; the layer now checks the panel image every 80 ms): size slider (panel width 15–80 cm, also in Options → Shot), opacity slider (click anywhere on the bar with the laser), attach to 🤚 left / ✋ right hand, 👤 head, 🌍 world, ✥ move (grip), ⟲ reset position, detection & buttons (auto/manual, shutter, type button). Same in the app's panel and in the background service
• 🪟 Panel: long translations are split into pages that fit – ▼ below the text shows the next page, ▲ goes back (1 / 3)
• 🪟 Panel: mode 🤖 automatic / ✋ manual right in the service list
• Round laser dot instead of the small square
• 🔧 Rebuild shows % – cargo's progress (Building … 120/231) becomes the percentage on the button; the progress lines stay out of the log
• ⚙ Works without the desktop app: new background service viewshot-daemon (Rust, daemon/). It does what the app did in the background – new photo → text recognition → translation → 🪟 VR panel, and 🔁 Lens with the overlay – so the app is mainly for settings and for looking at what you did (gallery, history). Same PaddleOCR models as RapidOCR via onnxruntime (identical text, boxes ±1 px), same services, same auto/manual routing, same translations.json cache and history – the app shows the service's results without asking again. Start: the layer sends hello to 127.0.0.1:47932 during a VR game and systemd (socket activation) starts the service; it quits ~2 min after the game (the OCR model leaves memory after 1 min idle). Options → Shot → Without the app: off / every VR game / only selected games (recently played games to tick + own names). While the app is open it holds ~/.local/state/linuxvr-viewshot/app.lock and the service steps aside. The app copies the service, onnxruntime and the OCR models to ~/.local/lib/linuxvr-viewshot/ and sets up the systemd user units (also AppImage / AUR). Custom snippet translator ("Eigene API") only with the app. Log: ~/.local/state/linuxvr-viewshot/daemon.log. daemon/contract.json (texts, defaults, cache keys, prompts) is generated from the app and checked by tests on both sides.
• ↻ Send again really asks again (the saved answer used to come back)
• 🪟 Translation panel in VR (Options → Shot, on by default): a window with the photo and numbers ①②③ on the recognised lines and the translations 1, 2, 3 below (AI answers as text). Service, task, from → to and ↻ right in VR: point with the controller (laser) and press the trigger – lists instead of dropdowns. Attached to the left/right hand, head or world. ✥ Edit mode (yellow border): grip = move, corner lights up + hold trigger: bottom right = width, top left = height; opacity − / + (30–100 %, also in Options); ⟲ reset position in the panel or in Options. The app draws the panel (invisible Qt window → panel/panel.png), clicks come back via UDP (port 47931). Grip/trigger still reach the game.
• 🥽 🔁 Lens: translation over the original (Options → Shot, off by default): a grey box with the translation over every recognised line inside the blue frame, updated with every new picture. Photos are shown in the 🪟 panel instead (photo overlay removed from the layer too, overlay_seconds is gone). The app draws the image (overlay/overlay.png), the layer shows it. If the number of lines doesn't match (e.g. AI answers) there is one big box over the text.
• 🖼 Local image LLM (Ollama) – new AI service: gets the photo plus the OCR lines, answers line by line (fits the VR overlay), fixes OCR mistakes and reads text OCR missed. Streaming, all AI tasks, model dropdown with the installed image models, 📥 Download model, and Keep in graphics memory (30 s … always) so VR gets the memory back. 📦 Install Ollama button: on Arch/CachyOS the pacman package that fits the graphics card (NVIDIA → ollama-cuda, AMD → ollama-rocm, else ollama) + systemctl enable --now ollama, on other distros the official script from ollama.com – runs in a terminal, hidden once Ollama is installed.
• Text recognition now also keeps the position of every line (cached)
• Answer question: the lines go to the AI numbered (1), 2), …) – the answer names the number (➜ ③ …), the same number as on the photo in the panel. Image LLM: first a short quiz? yes/no, then either a quiz prompt or a line-by-line translation (small models do two small steps much better than one big one)
• 🧭 Service per task (Options → Translation, one card): Mode 🤖 automatic / ✋ manual, Main, Explain context AI, Answer question AI (model = the one set for that AI, also changeable there). 🤖 Automatic (only with an AI as main – LibreTranslate & Co. switch the mode to manual): for every photo the main AI decides with a one-word question (QUIZ / EXPLAIN / TEXT), translates normal text itself and hands the rest to the task's AI. Picking a task on the main page / in the VR panel overrides it for the current + next photo, then it's automatic again. ✋ Manual: the task you pick stays, with its fixed AI
• 🏷 Last AI under Task (main page + VR panel): which AI, model and task wrote the answer shown; remembered per photo in the cache, stays until the next answer
• The service no longer switches automatically: if a service fails, Lingva/Google step in once and your setting stays
• Image LLM thinks first: for quizzes it first writes the full proverb/idiom/fact (💡 line), then picks the number – small models no longer just pick the first option as often

v0.4.6 rusti fix

rust clippi fix new version

v0.4.5

[v0.4.5] – 2026-09-28

• 🔁 Lens (live translation): the type trigger now works in automatic mode too – 🪄 Auto ↔ 🔁 Lens (manual: 🖼 → 📝 → 🔳 → 🔁 Lens). With 🔁 the shutter takes no photo but photographs the same area again every 1–10 s (Options → Shot; blue frame in VR, stays in front of your head) and the app translates it – fastest with LibreTranslate. Pictures go to live/live.png, not the gallery; unchanged text isn't translated again. Open a new frame to stop. New icons 🪄 Auto and 🔁 Lens. Needs a layer rebuild.
• AI answers appear live while Claude / Gemini / a custom command (e.g. Ollama) is still writing (ChatGPT/Codex usually delivers the answer in one piece)
• ☁/🔒 privacy note for every service (Options → Translation + tooltip in the Main dropdown): whether the recognised text is sent to the internet – only text, never the photo
• Tests: layer.json defaults in Python and Rust are compared automatically; new GitHub Actions workflow runs cargo test, clippy, pytest and the smoke test on every push
• Code: main_page.py split into main_translation.py, main_history.py, main_install.py, main_live.py (Lens) (+ ui/panel_window.py)

v0.4.4

[v0.4.3] – 2026-09-28

• ⭐ Favourite services (Options → Translation): tick services – then only those appear in the Main dropdown; nothing ticked = as before
• Main: 🕘 History pop-out – ⧉ opens the history in its own window (in VR its own panel); closing it docks it back
• Fix – AI task Answer question (ChatGPT only): with a blank (○○) plus answer buttons ChatGPT now names the WHOLE button to click (e.g. 下暗し), not just the missing character; its old saved answers are asked again
• Fix – Options → Shot: moving Frame size changed the Eye label instead of its own, and its Default button reset the eye slider
• Manual mode: icon corner selectable (Options → Shot): bottom left (new default) / bottom right / top left / top right – applies instantly in VR
• Fix – history missed photos: results that were already saved (e.g. text read earlier) now go into the history too; duplicates (same photo + same translation) are skipped
• Main: 🕘 History card above How it works (off by default – Options → Translation): scrollable list of past translations; tap one to show recognised text + translation (no image, the photo may be gone)
• Main: recognised text is editable – fix OCR mistakes or type text yourself, then Translate corrected text (the correction is remembered for that photo)
• Main: Recognised text can be folded (▸ / ▾), folded by default – more room for the translation in VR; the choice is remembered
• Main: 📋 Copy button right above the translation
• AI: Send again if it takes longer than … s (Options → Translation, per AI; off by default, 60 s): a stuck request is cancelled (whole process group) and sent again up to 2×, then Lingva/Google; off = wait up to 300 s
• Main: ↻ button next to the pop-out button – sends again and gets a fresh answer
• AI prompts: short answers, no tools / web search / file reading (Gemini sometimes took minutes); AI timeout 120 s → 60 s, a timeout says when the rate limit (429) was hit
• Main shows what the translation is doing right now (reading text / waiting for image detection / translating with X); new app log ~/.local/state/linuxvr-viewshot/ui.log with timings (Options → General → Folders)
• AI task Answer question: made for quiz / puzzle / escape rooms – names the button to click EXACTLY as written in the photo, or what to type into a gap (○○), first line starts with ➜; old saved answers are asked again
• Fix – Gemini "not running in a trusted directory" (error 55): the app now runs Gemini in trusted mode; a Gemini API-key login (settings.json / ~/.gemini/.env) counts as signed in; colour codes removed from error messages; the Sign-in terminal starts in the home folder
• Fix – "No text found" for minutes after start: at start the background detection reads ALL photos; the translation had to wait until it was done and showed "No text found" meanwhile. The translation now goes first, and "No text found" only appears after text recognition really ran
• Translation pop-out: ⧉ opens the translation in its own window (in VR its own panel); closing it docks it back
• QR codes inside the translation: links are clickable right in the translation field, 📋 copies – no extra box anymore
• AI task (Main, only with an AI service): 🌐 translate · 💡 explain context · ❓ answer a question in the photo (quiz, riddle …, otherwise translate)
• AI translation: Claude Code (Opus / Sonnet / Haiku), Gemini CLI, ChatGPT (Codex CLI) with model choice, plus a custom command (e.g. Ollama). Falls back to Lingva/Google if it fails
• Main: the service dropdown only lists services that are set up (key entered / program installed); model dropdown right next to it
• Models: Gemini flash / pro / flash-lite (CLI aliases, always the current model), ChatGPT gpt-5.6-luna (fast, default) / terra / sol / gpt-5.5; Codex gets the prompt via stdin and low reasoning effort, Gemini answers as JSON (no status lines in the translation)
• Model dropdown is a normal dropdown again (click opens the list); new model names via ✏ Other model …
• AI tools get a Sign in button (opens a terminal with claude / gemini / codex login) and show whether you are signed in; not signed in → skipped right away, Lingva takes over
• Claude Code is installed with the official installer (curl -fsSL https://claude.ai/install.sh | bash) instead of npm – the npm version could end up without its program ("claude native binary not installed"); a broken install is detected and offered as Reinstall. npm installs for Gemini/Codex now force optional deps + install scripts
• AI tools get an Install button (npm install -g --prefix ~/.local …, installs npm via pkexec if missing)
• Options → Translation: Test checks ONLY the chosen service (no Lingva/Google stand-in), and the service is never switched automatically while Options is open
• If the chosen service fails, Main switches the service to the one that actually translated (e.g. Claude Code → Lingva) and says why
• Fix – copy in WayVR didn't reach the desktop: copied text/images are also handed to the desktop clipboard via wl-copy (Options → General → Clipboard) – with a button that installs wl-clipboard (pacman / dnf / apt / zypper, password via pkexec)

v0.4.3

[v0.4.3] – 2026-09-27

• Fix – new VR photo not shown ("Take a photo in VR first"): the app noticed the photo while it was still being written. It now waits until the file is complete; text/QR detection no longer runs on half-written files
• VR layer saves photos atomically (hidden .part file → rename) – needs Rebuild & install once

v0.4.2 fix

[v0.4.2] – 2026-09-27

• AppImage (any distro, no Rust needed): the finished VR layer is inside; Install copies it to ~/.local, after an AppImage update it is refreshed automatically
• Fix – VR layer not loaded in VRChat with the AUR package: Steam/Proton games run in Steam's container and can't see /usr. The package no longer registers the layer system-wide; the app copies it to ~/.local (like the AppImage). Old system-wide registrations are detected and shown
• Fix: after Rebuild & install the button stayed disabled – Remove → Install works again
• Fix: text recognition crashed with newer omegaconf versions ("PosixPath is not a supported primitive type")

v0.4.1

[v0.4.1] – 2026-09-27

• Main: status, photo count and buttons in one compact row; new Remove button for the VR layer
• AUR package linuxvr-viewshot (system-wide); the app hides "Rebuild" when installed from a package and warns if the layer is installed twice
• Tests: pytest + smoke test; fixed a crash on exit in the smoke test (mouse-wheel filter is now removed before Qt shuts down)
• README with screenshots

[v0.4.0] – 2026-09-27

First public release.

VR (OpenXR layer)

• Hand-frame gesture: hold both grips → red frame → shutter trigger takes the photo
• Shutter on left / right / both triggers (default: right)
• Manual photo type: icon in the frame corner (🖼 / 📝 / 🔳), switched with a second trigger; stored in the PNG
• Frame size, eye (left/right) and excluded apps change live via layer.json

Desktop app

• Main: last photo with OCR + translation (Lingva, Google, LibreTranslate, DeepL, custom API), QR codes with open/copy
• Gallery: equal tiles, large view, copy / share / upload (directupload.eu) / info / delete, filter by type, bulk select → delete or set type
• Image detection: text / QR / image tags, VRChat nameplates don't count as text
• Clean up QR / text photos on exit (optional)
• WayVR: Cubee's theme + a watch button that opens this app inside VR
• App icon, correct name in task managers, EN/DE

Install

• install.sh for curl | bash (install / update / uninstall), AUR package linuxvr-viewshot