1. Start / Pause
Starts or pauses the translation process.
Full product guide
This page combines first-launch setup, interface explanation, OCR language setup, beta Audio Translation, Lens Mode, model selection, Custom OpenAI-compatible Local API, session recording, and optional Real-Time TTS in one place.
Thaluna.exe.Téléchargements, Pobrane, or names with accents.C:\Thaluna\ or D:\Apps\Thaluna\.Hardware note: on very new or partially unsupported GPUs, OCR and local translation can fall back to CPU. If you need stable behavior, open Settings → Performance and set both OCR Device and Translation Device to CPU. For faster translation, use Ollama on supported local GPU hardware or OpenRouter for cloud translation.
Display scaling note: if the OCR region appears offset, Windows display scaling is usually the cause. Set the relevant monitor to 100% scaling, keep the game and Thaluna on the same monitor if possible, and re-set the capture region after restarting Thaluna. Starting from v2.8, an experimental DPI scaling fix is available in settings.
Starts or pauses the translation process.
Opens Lens Mode.
Opens the translation backlog, similar to visual novel history.
Select the translation provider and target language.
Choose the language used for text detection.
Open the main Thaluna settings panel.
Hides the application background and leaves only subtitles visible.
Minimizes Thaluna to the taskbar.
Closes Thaluna.
Adjust the area where translated text appears.
Adjust the region Thaluna scans for text.
The screen area where Thaluna reads text from.
Built-in local translation models. Fast and lightweight, designed for weaker computers. Translation quality may be slightly lower compared to cloud models.
Allows you to use your own local models installed through Ollama.
Use a local OpenAI-compatible server such as llama.cpp, LM Studio, or another compatible endpoint.
Use cloud AI models through OpenRouter for higher quality translations.
Close the window without applying changes.
Confirm the selected model and apply the settings.
Displays the available translation models for the selected category.
Choose the translation model you want to use. If you select Ollama, your installed Ollama models appear here. If you select Local API, Thaluna uses the configured OpenAI-compatible endpoint.
Select the language you want Thaluna to translate the text into.
Apply the settings and start using the selected translation model.
Use when the source text is written in Chinese characters.
Use for Russian, Ukrainian, Bulgarian, and other Cyrillic-based text.
Recommended for English subtitles, UI text, and most Western PC workflows.
Use for Japanese games, visual novels, manga, and anime subtitles.
Use for Korean Hangul text.
Alternative recognition path for Latin-based languages, sometimes better for stylized fonts.
Opens the translation backlog with previously translated text.
Locks or unlocks the backlog window so you can move it freely.
Closes the backlog panel.
Scroll through previously translated lines.
Drag to increase or decrease the size of the backlog window.
Scans the entire screen and overlays translated text wherever text is detected.
Select a specific region of the screen to scan and translate. This often improves accuracy because OCR focuses on a smaller area.
Enable or disable the translation overlay displayed on the screen.
Close Lens Mode and return to the main Thaluna interface.
Improves text formatting for manga panels and speech bubbles.
Improves detection and formatting for vertical Japanese text.
Choose the language of the text you want Thaluna to detect.
Adjust the size of the translated text displayed on the screen.
Quickly hide the Lens interface when you want only the translated overlay visible.
Continuous Lens removes the need to manually re-select the same area every time the content changes. It is especially useful for games, videos, and dynamic UI regions where the text updates in place.
Choose a predefined theme for the real-time interface.
When Theme = Custom, adjust text color, background color, and UI/frame color for real-time mode.
Assign keys for Minimal Mode, hide icon, toggle mode, snapshots, overlay toggle, and preset loading.
Click an empty field and press a key such as A, B, J, or F5 to assign it.
Adjust the translated subtitle size in real-time mode.
Enable or disable the animated star background effect in the translation window.
Increase outline thickness to improve readability over bright or busy backgrounds.
Change text color, background color, and background opacity specifically for Snapshot / Lens Mode overlays.
Set your OCR and translation windows where you want them, give the preset a name, and click Save.
Use Load to restore a saved layout instantly for the current game, video player, or monitor setup.
Presets are especially useful when you switch between different games, monitors, or media players.
Standard works for most cases, Pixel Text improves game-font detection, and Manga mode targets harder vertical Japanese layouts in real-time mode.
Enable auto-crop when the capture area is larger than the subtitle line and empty margins make OCR less reliable. Disable it for bright visual novel backgrounds or carefully tuned manual regions.
Reduces OCR refresh pressure for weaker systems or when you are running other heavy applications.
Pauses OCR after static text remains unchanged for a while, helping reduce unnecessary CPU usage.
Choose Auto, CPU, or GPU for OCR recognition.
Choose Auto, CPU, or GPU for built-in local translation models. This does not affect Ollama or OpenRouter, which run externally.
Audio Translation lets Real-Time mode use live system audio transcription instead of screen OCR. It is useful for anime, videos, streams, voice-acted visual novels, and situations where subtitles are unavailable or hard for OCR to read.
Beta note: speech recognition quality depends on source audio, background music, overlapping voices, source language, selected STT model, translation backend, and audio device configuration.
Select a proper loopback device such as Speakers or Headphones loopback so Thaluna can hear the application audio.
Audio Translation uses OpenRouter Speech-to-Text, so you need a valid OpenRouter API key, available credits, and a supported STT model.
Choose the source language when possible. Auto-detection can work, but explicit language selection is usually more stable for anime and voiced content.
Use the Stable preset first. Lower-latency modes can feel faster, but they may produce more unstable partial transcripts.
If audio activity is detected but no text appears, raise the source application or system volume and verify that the selected loopback device is the one actually playing audio.
The STT step only creates source text. The final translation still uses your selected backend: built-in models, Ollama, Custom Local API, or OpenRouter translation.
Refresh the list of available local Ollama models after starting Ollama or downloading new ones.
Set the base URL for an OpenAI-compatible local server, usually http://127.0.0.1:8080/v1, and keep the model ID matching your server setup.
Only fill this if your local endpoint requires one. Most localhost llama.cpp setups can leave it empty.
Paste your OpenRouter API key carefully, with no extra spaces, and keep it private.
Choose the OpenRouter model used for translation. Models differ in speed, cost, and quality.
Manually specify a model ID when you want to use something not listed in the dropdown.
Instant gives the fastest cloud responsiveness, while Economy reduces request frequency to save tokens.
Adjust request timing, ignore very short text, and skip highly similar OCR lines to reduce unnecessary API usage.
http://127.0.0.1:8080/v14096-81920.1-0.3512-2048Example: llama-server -m model.gguf --port 8080 --ctx-size 8192
Thaluna sends short OCR translation requests, so very large context windows are usually not needed. If a model generates very long nonsense responses, lower temperature or set a stricter max token limit in your local server/client configuration.
Enable or disable automatic transcript recording for the current session.
Choose where the TXT transcript files should be written.
Choose between translation-only output or original-plus-translation output.
Repeated duplicate lines are filtered to keep the transcript readable over long sessions.
Real-Time TTS is optional and does not affect the rest of Thaluna if you leave it disabled.
Choose whether TTS reads translated text or the original OCR text.
Select a supported voice from the list and adjust speech speed to fit your preference.
TTS uses your configured OpenRouter API key. Audio is AI-generated, and you are responsible for the text you choose to send.
If you only need one specific part of the workflow, use the dedicated guides for Local API setup, OpenRouter setup, display scaling, hardware compatibility, or the changelog.