Hey!

First off — genuinely, thank you. Not just for the idea itself, but for taking the time to explain it properly. That’s not something everyone does, and I really do appreciate it.

And just to be clear — you didn’t sound arrogant at all. You sounded like someone who actually uses the tool and knows what’s missing. That’s exactly the kind of feedback that’s worth something.

What you’re describing is basically a hybrid between Real-Time Mode and Lens Mode — you define the area once, and then the app watches it on its own, detects when something changes, and refreshes the overlay automatically without needing a new manual snapshot each time. I get why that would be useful, especially for manga, webtoons, or UI-heavy games.

Your technical point about local change detection also makes sense — triggering OCR and translation only when the area actually changes is the right way to do it, so you’re not burning API calls for nothing.

So yes — I’m taking this seriously. But I’ll be straight with you: it’s not a small change. This would be its own separate workflow, not just a tweak to Lens Mode, so I can’t give you an ETA right now.

For the time being, the two options stay the same:

- Real-Time Mode — continuous OCR inside a defined capture area

- Lens Mode — snapshot-based overlay, great when layout matters

But the “real-time overlay with a remembered area” idea is on the list now. And I’m glad it is.

Thanks again.

Hey,

Thanks a lot for the detailed report — I really appreciate it.

For manga reading, Lens Mode works best with Area Selection rather than Full Screen Scan.

If you scan the entire screen, it can also pick up UI elements like the taskbar, clock, or browser UI, so for manga readers it’s best to select only the page area you want to translate.

Also, Lens / Manga Mode currently works as a snapshot-based workflow.

It does not automatically detect page turns or scrolling, so after changing the page you need to trigger a new snapshot.

If you want a workflow that updates automatically without taking a new snapshot every time, the better option is to use the normal Real-Time Mode and move the OCR area over the bubbles / text as you read.

The Economy Mode / similarity / short text settings mainly affect cloud translation request behavior, but they do not make Lens Mode auto-refresh on page changes.

For English manga, “Universal (Latin)” is the correct OCR option in Lens Mode — it is intended to cover English / Latin-script text as well.

If the selected area is still being ignored even when using Area Selection correctly, please send:

- whether you are using Full Screen Scan or Area Selection

- whether it happens every time or only sometimes

- which translation source/model you are using

That would help determine whether there is a real capture-region bug.

Thanks again for the feedback.

Also I made a quick visual guide to make things clearer.

1 — This button lets you select a specific area on the screen.  

For manga, this is the recommended option.  

It’s better to use this instead of the full screen snapshot, because scanning the whole screen can also pick up UI elements like the taskbar or browser.

2 — This is Manga Mode.  

It formats the translated text so it fits inside manga bubbles instead of showing plain text.

3 — Vertical Text.  

You can enable this if you want the text to be displayed vertically (useful for some manga layouts).

4 — Target language.  

Here you choose the language you want to translate into.

OCR Language can stay on “Universal (Latin)” — it works well for English and is designed to support Latin-based text in general.

Below I included an example using your screenshot so you can see how it should look in practice.

Also, the bubbles that appear can be adjusted:

- you can move them

- resize them with the mouse wheel

- remove them with right click

This is the intended manga-reading workflow for Lens Mode.

Also, if you are using Cloud Mode and run into rate-limit or provider errors, try switching to a different OpenRouter model/provider.

Some providers can be temporarily more restrictive or unstable than others, even when the rest of the workflow is working normally.

Hey! Thanks for the kind words, and that's a really good question honestly.

Your reasoning makes total sense — if Lens Mode works fine with 125% scaling, it's natural to assume the capture window would too. The thing is, they actually use different capture paths internally, which is why one works and the other doesn't (yet).

The core issue isn't screen capture itself — it's how the app currently handles DPI and coordinate mapping. At 100% scaling everything lines up cleanly, but at 125% or 150% Windows and Qt can report coordinates differently (logical vs physical pixels), which causes the captured region to get offset even if the window looks perfectly fine visually.

It's something we'd like to fix, but it's not a straightforward change — the tricky part is making sure it works correctly across different scaling values, single and multi-monitor setups, mixed scaling across monitors, and different display modes. So it needs proper testing before we can ship it reliably.

For now, 100% scaling is the safest workaround for the capture window. Sorry for the inconvenience, and thanks for taking the time to explain your setup so clearly!

Sorry for the late reply — your post got caught in the spam filter and I wasn't notified about it.

Yeah, it should work fine even without a dedicated GPU.

OCR runs well on CPU, so that part's no issue. For translation though, running local models without a GPU will be pretty slow — so I'd recommend using OpenRouter as the translation device. That way OCR still runs locally on your CPU, and OpenRouter handles the translation part.

Feel free to try the demo first to see if everything runs on your laptop — just keep in mind it uses local models, so translation will be slower there. If it works, setting up OpenRouter is the way to go. There's a guide on the forum — costs are really low, like $5 can last for months.

It looks like the language was set in the wrong place.

The Target Language is the language you want the text to be translated into.

The OCR Language is the language the text is originally written in.

In your case the video text is Chinese, so try this:

OCR Language → Chinese
Target Language → English

Right now it looks like Chinese is selected as the target language, which means the translator is trying to translate into Chinese instead of translating the Chinese text to English.

After changing the OCR language, restart the capture and it should work correctly.

This screenshot shows Japanese as an example, but in your case set it to Chinese.
The rule is the same.

You're welcome! I'm glad we got it working.

If you ever have questions about the settings, I also wrote a complete guide for Thaluna on the forum that explains the different options.

And if something isn't covered there, you can always post here again and I'll help when I can.

Yes, correct. Just download the first one (Windows x86_64)

Hello!

No problem at all, I'm glad it's working for you now.

To update Thaluna, just go to the main download page:

https://thalunadev.itch.io/thaluna-real-time-translator

Then simply download the Windows version again and replace your old Thaluna folder with the new one.

That's all you need to do to update.

Hello,

Great to hear that OpenRouter is finally working for you!

From the screenshot it looks like the issue is related to the OCR language setting

Right now the OCR language is set to English, but the game/movie text you are trying to translate appears to be Chinese or Japanese.

Because of this, the OCR cannot correctly read the characters, so the translator only receives small fragments like "R." instead of the full sentence.

Please try this:

1. Go to the OCR language setting

2. Change it from English to the correct language (Chinese or Japanese depending on the game/movie)

3. Save the settings and try again

Once the OCR language matches the game text, the full sentence should be detected and translated correctly.

Also, I noticed from the screenshot that you are still using an older version of Thaluna.

A newer version is available which includes fixes and some improvements. If you want, you can update it anytime by downloading the latest version from the Thaluna page.

Please note that some visual settings (like colors) may reset because they are stored in the config file, but your OpenRouter API key should remain saved since it is stored separately.

Hello

Please check the following:

1. Generate a new API key on OpenRouter.

2. In Thaluna settings, delete the old key completely and paste the new one.
   Make sure you paste the entire API key exactly as OpenRouter gives it, not just part of it.

3. Click the button next to the API field to reveal the key and verify it looks correct.

4. 
   

5. Click Save / OK, then fully restart Thaluna.

If the problem still happens, please check the stored configuration manually:

1. Press Win + R

2. 

3. Type:

%localappdata%\Thaluna\secrets

1. Open openrouter_api_key.json

2. Verify that the API key stored there matches your current OpenRouter key exactly.

This error usually means one of the following:

• the API key is missing, incorrect, or saved incorrectly

• the app is still using an old/invalid key

If the steps above still don't fix the issue, please try one quick test:

Temporarily switch to a different internet connection (for example your phone hotspot) and test Thaluna again.

Sometimes a VPN, proxy, firewall, or antivirus web protection can interfere with API authorization requests.

Using a different network for a quick test can help confirm if the issue is network-related.

Thaluna Interface Overview

1️⃣ Start / Pause
Starts or pauses the translation process.

2️⃣ Lens Mode
Opens Lens Mode

3️⃣ Backlog Viewer
Opens the translation backlog (similar to visual novel history).

4️⃣ Translation Language
Select the translation provider and target language.

5️⃣ OCR Language
Choose the language used for text detection.

6️⃣ Application Settings
Open the main Thaluna settings panel.

7️⃣ Minimal Mode
Hides the application background and leaves only subtitles visible.

8️⃣ Minimize Window
Minimizes Thaluna to the taskbar.

9️⃣ Close Application
Closes Thaluna.

🔟 Resize Text Display Area
Adjust the area where translated text appears.

1️⃣1️⃣ Resize OCR Capture Area
Adjust the region Thaluna scans for text.

1️⃣2️⃣ OCR Capture Region
The screen area where Thaluna reads text from.

Translation Model Selection

1️⃣ Standard Models
Built-in local translation models. Fast and lightweight, designed for weaker computers. Translation quality may be slightly lower compared to cloud models.

2️⃣ Ollama Models
Allows you to use your own local models installed through Ollama.

3️⃣ OpenRouter (Cloud)
Use cloud AI models through OpenRouter for higher quality translations.

4️⃣ Cancel
Close the window without applying changes.

5️⃣ OK
Confirm the selected model and apply the settings.

6️⃣ Model List
Displays the available translation models for the selected category.

How to Select a Translation Model

1️⃣ Select a model
Choose the translation model you want to use.
If you select Ollama, your locally installed Ollama models will appear here.

2️⃣ Choose Target Language
Select the language you want Thaluna to translate the text into.

3️⃣ Click OK
Apply the settings and start using the selected translation model.

Selecting OCR Language

The OCR Language tells Thaluna what language it should detect on your screen before translating it.

1️⃣ Choose the OCR language
Select the language that appears on your screen.
For example, if the game or video text is Japanese, select Japanese.

2️⃣ Click OK
Confirm your selection.

Available OCR Languages

Chinese
Use this when the text on your screen is written in Chinese characters.

Cyrillic
Used for languages written in the Cyrillic alphabet (Russian, Ukrainian, Bulgarian, etc.).

English (Default)
Recommended for English text.

Japanese
Use this for Japanese games, visual novels, anime subtitles, etc.

Korean
Use this when detecting Korean Hangul text.

Latin (Alternative)
Alternative recognition for Latin-based languages (English, Spanish, French, German, Polish, etc.).
Sometimes works better with stylized fonts.

Backlog Panel

The Backlog shows previously translated lines, similar to the dialogue history in visual novels.

1️⃣ Backlog Button
Opens the translation backlog with previously translated text.

2️⃣ Lock / Unlock Backlog
Locks or unlocks the backlog window so you can move it freely.

3️⃣ Close Backlog
Closes the backlog panel.

4️⃣ Scroll Bar
Scroll through previously translated lines.

5️⃣ Resize Handle
Drag to increase or decrease the size of the backlog window.

Lens Mode

Lens Mode is a screen overlay that detects text on the screen and places the translated text directly over the original text.

1️⃣ Full Screen Scan
Scans the entire screen and overlays translated text wherever text is detected.

2️⃣ Area Selection
Select a specific region of the screen to scan and translate.
This often provides better accuracy, because the OCR focuses on a smaller area.

3️⃣ Toggle Translation Overlay
Enable or disable the translation overlay displayed on the screen.

4️⃣ Exit Lens Mode
Close Lens Mode and return to the main Thaluna interface.

5️⃣ Manga Mode
Improves text formatting for manga panels and speech bubbles.

6️⃣ Vertical Text
Improves detection and formatting for vertical Japanese text.

7️⃣ OCR Language
Choose the language of the text you want Thaluna to detect.

8️⃣ Font Size
Adjust the size of the translated text displayed on the screen.

General Settings

Theme Style

Theme
Select a predefined theme for the real-time mode interface.
Each theme contains a different color palette for the UI and translated subtitles.

Custom Colors
This section works only if Theme = Custom.

• Text Color – changes the color of translated text.

• Background Color – changes the background color behind the subtitles.

• UI/Frame Color – changes the color of UI elements (buttons, frames, etc.).

These changes apply only to Real-Time Mode.

Shortcuts

Global keyboard shortcuts that allow you to control Thaluna without clicking the interface.

To set a shortcut, click the empty field and press a key (for example A, B, J, F5, etc.).
After that, the function will be triggered by that key.

Available shortcuts:

Minimal Mode
Hides the background in real-time mode so only subtitles remain visible.

Hide Icon
Hides the Thaluna icon in real-time mode.
It is recommended to keep the icon visible so you can easily grab the window or disable Thaluna if needed.

Toggle Mode
Switches between Real-Time Mode and Lens Mode.

Quick Snapshot (Full Screen)
Starts Lens Mode and captures the entire screen.

Quick Snapshot (Select Area)
Starts Lens Mode and allows you to select an area of the screen.

Snapshot: Toggle Overlay
Shows or hides the translation overlay in Lens Mode.

Load Preset 1 / 2 / 3
Loads saved window position presets for Thaluna (subtitle window placement on screen).

Appearance

Font

Translation Font Size
Adjusts the size of translated subtitles in Real-Time Mode.

Effects / Details

Show Stars Effect in Translation Window
Enables or disables the animated star background effect in the translation window (Real-Time Mode).

Text Outline Thickness
Controls the thickness of the black outline around translated text.
A stronger outline can improve readability on bright or complex backgrounds.

Snapshot Mode Appearance

These settings affect the appearance of Lens Mode (overlay translation mode).

Change Text Color
Changes the color of translated text displayed on the screen overlay.

Change Background Color
Changes the background color behind translated text in Lens Mode.

Window Position Presets

Presets allow you to save and quickly restore the position of Thaluna windows on the screen.

This includes:

• the OCR capture area (where Thaluna reads text from the screen)

• the translation window position (where translated text appears)

How to use presets

1. Adjust the OCR capture area and translation window position where you want them on the screen.

2. Enter a name for the preset in one of the preset fields.

3. Click Save.

The current layout will now be saved.

Loading a preset

Click Load next to a preset to instantly restore the saved window layout.

Thaluna will move the windows back to the exact positions that were saved.

When presets are useful

Presets are especially useful when you:

• use different games or video players

• have different screen layouts

• want to quickly switch between setups

They also work very well together with keyboard shortcuts, allowing you to load layouts instantly.

Performance & Modes

OCR Mode

Changes how the OCR engine detects text on the screen.

Standard (Default, for clear text)
Recommended for most situations. Works well for normal subtitles, UI text, and most fonts.

Pixel Text (Games, stylized fonts)
Improves detection for pixel-style or heavily stylized fonts often used in games.

Manga (Vertical text detection)
Designed to better detect vertical Japanese text in Real-Time Mode.

Note:
The default Standard mode already supports vertical Japanese text, but this mode can sometimes give better results.

This mode only affects Real-Time Mode.

Low CPU Mode

Reduces CPU usage by lowering how often OCR refreshes the screen.

This means translation updates slightly slower, but uses fewer system resources.

Useful if:

• your CPU usage is high

• you are running other heavy programs

Enable Auto-Pause

If the detected text does not change for 10 seconds, Thaluna will automatically pause processing.

You can resume processing manually using the pause/play button in the UI.

This helps reduce unnecessary CPU usage when the screen is static.

Hardware Acceleration

OCR Device

Select which hardware runs the OCR detection.

Auto
Automatically selects the best available option.

CPU
Most compatible option but slower.

GPU
Uses GPU acceleration for faster OCR.

Currently works mainly with NVIDIA GPUs that support CUDA.

Note:
Some very new GPU architectures may not yet be supported by the OCR backend.

Translation Device

Select which hardware processes the translation model.

CPU
Works on any system but is much slower.

GPU
Much faster for local AI models.

Important:
This setting only matters when using local translation models.

If you use:

• Ollama

• OpenRouter

then translation is processed externally and this setting has no effect.

Ollama / Cloud Settings

This section allows you to configure local Ollama models and cloud translation through OpenRouter.

Refresh Ollama Models

Refresh Ollama Models

Updates the list of available models from your local Ollama installation.

Use this if:

• Ollama is running

• you downloaded new models while Thaluna was already open

Clicking Refresh will scan Ollama again and display any newly installed models.

OpenRouter (Cloud Translation)

These settings allow Thaluna to use cloud-based AI translation models through OpenRouter.

API Key

Paste your OpenRouter API key here.

Important:

• Do not add spaces before or after the key

• Keep your key private

• Never share it publicly

Model

Select which OpenRouter model will be used for translation.

Different models may vary in:

• speed

• cost

• translation quality

The default model is chosen to be fast and inexpensive.

Custom ID

Allows you to manually specify a custom OpenRouter model name.

Use this only if you want to use a specific model that is not listed in the dropdown.

Update Mode

Controls how often translation requests are sent to the cloud model.

Instant (higher token usage)
Translates every detected change immediately.
Best responsiveness but may use more tokens.

Economy
More conservative mode that reduces request frequency to save tokens.

Additional economy options allow you to fine-tune behavior.

Economy Settings

These options control how often translation requests are sent.

Minimum time between requests
Defines the minimum delay between translation requests.

Ignore text shorter than
Very short text can be ignored to avoid unnecessary requests.

Similarity skip threshold
If the detected text is very similar to the previous one, translation can be skipped.

This helps reduce unnecessary API usage.

Session Recording

This feature allows Thaluna to automatically save a transcript of translated text during a session.

The transcript is saved as a TXT file while the application is running.

Repeated duplicate lines are automatically filtered to keep the transcript readable.

Save session transcript automatically

Enable or disable automatic recording of translated text.

When enabled, Thaluna will continuously save translations to a text file during the session.

Save folder

Select the folder where the transcript files will be stored.

You can:

• Browse to choose a different location

• Open Folder to quickly access the current save directory

Save mode

Choose what information should be saved in the transcript.

Translation only (recommended)
Saves only the translated text.

Original + translation
Saves both the original detected text and its translation.

This error means OpenRouter rejected authentication (401: Missing Authentication header).

Usually this happens when the OpenRouter API key is missing or incorrect in the app settings.

Please try the following:

1. Go to Settings → Ollama / Cloud.

2. Paste your OpenRouter API key again (make sure there are no spaces before or after the key).

3. Save the settings and restart Thaluna.

4. If it still doesn't work, generate a new API key on OpenRouter and paste it again.

If you are using a VPN, proxy, or firewall, try disabling it temporarily and test again.

When OpenRouter authentication fails, Thaluna automatically switches to a local fallback translation model.

If you're still unsure whether the API key was saved correctly, you can also check it manually.

Sometimes when copying a key (especially after regenerating it or changing limits), a space or extra character can accidentally be included. That can cause the authentication error.

Thaluna stores the OpenRouter key in the following file:

C:\Users\YourUserName\AppData\Local\Thaluna\secrets\openrouter_api_key.json

The easiest way to open it:

1. Press Win + R on your keyboard.

2. Type:

%localappdata%\Thaluna\secrets

1. Press Enter.

2. Open the file openrouter_api_key.json with Notepad and check if your API key is there.

You can also paste your key into this file if needed.

⚠️ Make sure Thaluna is completely closed before editing the file, otherwise the change may not save correctly.

This method can help confirm that the correct key is stored without any extra spaces or characters.

Also remember to keep your API key private and never share it publicly.

Hey! You're very close — the API key part looks correct 👍

I made a quick visual guide for you (see the screenshots). Just follow these steps:

1️⃣ Click the Language button at the top of Thaluna.

2️⃣ In the window that opens, switch to the OpenRouter (Cloud) tab.

3️⃣ Select OpenRouter: Cloud Translation.

4️⃣ Choose your Target Language (for example English).

5️⃣ Click OK.

After that Thaluna will start translating using OpenRouter.

One small recommendation:
I noticed your API key currently has no spending limit (unlimited).

OpenRouter is prepaid, so it will only use your balance, but it's still a good idea to set a small limit (for example $5) on the key.

This just prevents accidental higher usage if something runs in the background.

Hey! Thanks for the screenshots — that helps a lot 

The "Ollama server not found" message is completely normal if you're not using Ollama. You can safely ignore it — OpenRouter does not require Ollama at all.

Since your system doesn't allow GPU selection, it most likely means you don't have a CUDA-capable NVIDIA GPU. In that case everything runs on CPU:

- OCR → CPU

- Local translation models → CPU

That's why it can struggle to keep up in visual novels or videos — your CPU is handling both tasks at the same time.

Using OpenRouter will help because your CPU will only handle OCR, while translation runs in the cloud instead.

Here's how to set it up:

1. Go to https://openrouter.ai and create an account.

2. Add a small balance (even $5 lasts a very long time).

3. Generate an API key.

4. In Thaluna → Settings → Ollama/Cloud section, paste your API key in the OpenRouter field.

⚠️ Important: Do NOT share your API key publicly or paste it in chat. Keep it private.

5. Leave the default model (Gemini 2.5 Flash Lite).

6. Go to Languages → select OpenRouter as the translation provider.

7. Choose your target language and click OK.

Also make sure the video or game is not running in true fullscreen mode — if it takes over your whole screen, try switching to Windowed or Borderless Windowed in the display settings.

Try that and let me know how it goes 

Hey! Thanks for reaching out — and no worries, we’ll figure it out 👍

If the translation can’t keep up in visual novels, it’s usually related to performance settings. Thaluna actually runs two separate things at the same time:

• OCR (reading the text from the screen)
• Translation (processing that text)

If both run on CPU on a laptop, it can get overwhelmed.

Here’s what I’d suggest trying:

1. Make sure you’re using the latest version — recent updates improved OCR behavior and device switching quite a bit.

2. Set OCR device to GPU (if your GPU supports it).
You can find this in Settings → Performance.
OCR on CPU can slow things down a lot on weaker systems.

You can also try Pixel Text mode — sometimes it works better with VN fonts.

3. If you're using built-in local translation models, they also use CPU/GPU resources.
On lower-end laptops, running both OCR and translation locally can cause lag.

For smoother performance, a good setup is:

• OCR → GPU
• Translation → OpenRouter (Cloud Mode)

That way your PC only handles OCR, and translation runs externally — which usually makes a noticeable difference on weaker hardware.

There’s a simple OpenRouter guide in the Forum if you want to try it. And don’t worry about cost — Gemini Flash Lite is extremely cheap, so even $5 lasts a long time.

Also make sure the game is running in windowed or borderless mode — exclusive fullscreen can block screen capture updates.

If you’d rather stay fully local, tell me your CPU/GPU/RAM and I can suggest a lighter model that might work better 👍

As for translations stopping or looking weird at first — this can happen if the models are still loading when you start. Give it a few seconds after launching before playing. It's also worth double-checking that the OCR language matches what's on screen. And updating to the latest version should help too — we fixed a bug where settings sometimes didn't apply correctly.

Hi!

The error you saw (WinError 1455) is related to Windows memory, not your Thaluna settings.

Sometimes Windows does not allow enough memory for the AI engine to start, especially right after boot or if many programs are running in the background.

Please try this first:

1. Close Thaluna

2. Restart your laptop

3. After restart, wait a minute and do not open other apps yet

4. Start Thaluna first

Many laptops automatically start apps in the background (Discord, browser, game launchers, cloud sync, etc.), and that can use memory before Thaluna starts.

If it still happens, please check this Windows setting:

1. Press Windows key + R

2. Type: sysdm.cpl

3. Press Enter

4. Go to the Advanced tab

5. Under Performance, click Settings

6. Go to the Advanced tab again

7. Under Virtual memory, click Change

8. Make sure Automatically manage paging file size for all drives is enabled

9. Click OK and restart your PC

This allows Windows to automatically manage memory for larger applications like AI tools.

Also, please make sure your C: drive has some free space (at least a few GB), as Windows needs space for memory management.

If anything looks different on your screen, just send me a screenshot and I’ll guide you step by step

Hi! Thank you again for the screenshots 

I can see in your screenshot that scaling is still set to 175%. That means Windows is still using 175% internally.

When Custom Scaling is enabled, Windows requires you to sign out or restart before the new value fully applies. Until you do that, it will continue behaving like 175%.

Please try this carefully:

1. Go to Settings → System → Display
2. Under Scale, make sure it says exactly 100% (Recommended)
3. If you see “Custom scaling is set” at the top, click Turn off custom scaling and sign out
4. Fully sign out or restart your laptop
5. After logging back in, check again that Scale shows 100%
6. Then open Thaluna and reset the capture region

Right now the misalignment is happening because Windows is still running at 175% scaling.

Once it truly switches to 100%, the capture region should align correctly.

About “Windowed / Borderless Windowed”:

This is not a Thaluna setting.

It refers to how the video is displayed.
If possible, keep the video in a normal window instead of exclusive fullscreen mode.

Let me know what happens after the full restart 

Hey!

I think I found what's going on. I was actually able to reproduce the exact same issue on my end.

It looks like this is a Windows display scaling DPI problem. When scaling is set above 100%, Thaluna's capture box visually appears in the right place, but internally it reads from a slightly different region of the screen. That is why the translation engine is working correctly, but the OCR region is slightly offset.

Could you try a few things:

1. Set Windows display scaling to 100% for the monitor you're playing on. You can change it back after testing.

2. If you have multiple monitors, either set both to 100% or move the game and Thaluna to the same monitor.

3. Run the game in Windowed or Borderless Windowed mode instead of exclusive fullscreen.

4. After changing the scaling, restart Thaluna and re-set the capture region.

If it still does not work after that, try switching the OCR device to CPU in settings just to rule out anything GPU related.

If the issue persists, let me know your Windows scaling percentage, number of monitors, game resolution, and OCR settings. That will help narrow it down further.

I am also working on improving DPI and scaling handling in an upcoming update, so this should not be a permanent workaround.

✅ Step 1 – Create an OpenRouter Account

1. Go to: https://openrouter.ai
2. Create an account
3. Add a small balance (recommended: $5 for testing)

🔑 Step 2 – Generate an API Key

1. Open your OpenRouter dashboard

2. Go to API Keys

3. Click Create New Key

4. Copy the generated key

Keep this key private.

⚙️ Step 3 – Add the Key in Thaluna

1. Open Thaluna

2. Go to Settings → Ollama/Cloud

3. Paste your API key

4. Select a model preset (or enter a custom model ID)

You are now ready to use Cloud Mode.

🌍 Step 4 – Enable OpenRouter in Language Settings

After setting your API key and model:

1. Go to Languages

2. 

3. In the window that opens, switch to the OpenRouter (Cloud) tab.
4. Select OpenRouter as your translation source
   

5. 

6. Select OpenRouter as your translation source
   

7. Choose your Target Language (for example English)
   

8. 
   

9. Click OK

After that Thaluna will start translating using OpenRouter.

💡 Recommended for Real-Time Use

For best balance of speed and cost:

• Fast & Cheap → Gemini Flash Lite

• Balanced → GPT-4o Mini

• Multilingual Strength → DeepSeek

(You can also enter any custom model ID supported by OpenRouter.)

💰 Cost Example

Cloud translation is extremely affordable for normal real-time use.

For typical real-time text (games/anime), costs are usually only a few cents per hour depending on model and usage.

👉 Always check your OpenRouter dashboard for live pricing.

🔒 Security

• Your API key is stored locally on your system (per-user).

• Thaluna does not host or distribute any third-party models.

• All usage is performed through your own OpenRouter account.

🛠 Troubleshooting

If Cloud Mode is not working:

• Check your API key
• Make sure you have internet access
• Verify your OpenRouter balance
• Try switching to a different model

If issues persist, post logs in this forum section.

To disable Cloud Mode at any time, simply switch back to Local Mode (built-in or Ollama).

## RTX 5000 Series (Blackwell GPUs)

Due to upstream framework limitations, GPU acceleration for built-in OCR

(PaddleOCR / MangaOCR) is not yet available on RTX 5000 series GPUs.

### Recommended setup:

• OCR Device → CPU
• Translation Device → CPU
• Use Ollama for GPU-accelerated translation (recommended)
• Or use OpenRouter (cloud) as an alternative provider  

Ollama uses its own runtime and works correctly on RTX 5000 GPUs.

OpenRouter runs in the cloud and does not rely on your GPU.

---

## Non-ASCII Windows Usernames (Cyrillic, Arabic, etc.)

Thaluna now automatically detects non-ASCII usernames and switches to:

C:\Users\Public\Thaluna

No manual action is required.

---

## Diagnostics

If you experience crashes or OCR issues:

• Use the **Diagnostic build**
• Attach console output or logs when reporting issues

Hi! Honestly, those logs were a goldmine, thank you so much for sending them. We finally caught the culprit.

Good news first: the path issue is officially dead. I could see in the logs that Thaluna successfully switched to the safe path: C:\Users\Public\Thaluna. So the Cyrillic username problem is gone for good 👍

The only thing left is your GPU. Since the RTX 5060 is so brand new (Blackwell), the current OCR engine just doesn't know how to talk to it yet. That's why you're seeing those "fatal kernel" errors , it's trying to use tools that don't exist for your card.

Here’s how to get it working right now:

1. Grab the latest Diagnostic build I just uploaded to Itch This one includes an additional fix:
   Manga Mode now correctly respects the OCR Device setting and will stay on CPU instead of trying to use CUDA automatically. This prevents unnecessary GPU initialization on very new GPUs like the RTX 5060.
2. Force it to CPU: Go into Settings -> Performance and set both OCR Device and Translation Device to CPU.
3. Save and restart.

If the UI gives you any trouble, you can just open Thaluna_internal\config.json and manually change both devices to "cpu".

Since CPU translation can be a bit slow, I really recommend using Ollama. Your 5060 is a beast, and Ollama will use it perfectly for translation, even if the OCR is stuck on the CPU for now. I made a quick guide on how to set it up here:

Hi!
Thank you so much for the detailed logs — they were extremely helpful. We finally pinpointed what’s going on.

There are two separate issues happening at the same time:

1. Non-ASCII Windows username (primary cause)
Your Windows username contains Cyrillic characters (Антон). Unfortunately, PaddleOCR has a known issue with non-ASCII usernames on Windows. The path gets corrupted internally (Антон → AHTOH), which causes PaddleOCR to fail when loading its models.

To fix this, I’ve updated Thaluna to use a safe system path that does not include the username:
C:\Users\Public\Thaluna\
This completely avoids the path corruption problem.

2. RTX 5060 (Blackwell) GPU — too new for PaddleOCR
Your RTX 5060 uses the Blackwell architecture (sm_120). The current PaddleOCR / CUDA build does not support this GPU yet, which is why you see repeated sm_120 / kernel errors. This is a limitation of the OCR engine, not your system.

How to run Thaluna successfully right now

1. Download the Diagnostic build again
   I’ve just uploaded a newly updated Diagnostic build that includes the username path fix.
   Please re-download the Diagnostic version from the same Itch page (older diagnostic builds won’t include this fix).
2. Set OCR Device to CPU
   Go to Settings → OCR Device → CPU
   GPU OCR will not work on RTX 5060 yet.
3. Translation Device behavior (important)

• If you use built-in translation, it must also run on CPU
• If you want fast translations, use Ollama instead

1. Use Ollama for fast translation (recommended)
   Ollama works perfectly on modern GPUs like RTX 5060.
   I recommend using the Gemma3:4b model.

Step-by-step guide here:
https://www.youtube.com/watch?v=_9PjiXKfxUY

After downloading the updated Diagnostic build and switching OCR to CPU, please try starting the app again.

If anything still fails:

• Let the Diagnostic version open (CMD window will appear)
• Copy the console output
• Paste it here

Thanks again for your patience — your report genuinely helped improve Thaluna for users with newer hardware and non-ASCII usernames.

Hi!

No worries at all 👍

The Diagnostic build is on the same download page, just listed as an additional file. You don’t need a separate link.

Here’s how to find it:

1. Go to the Thaluna download page (where you normally download the app).

2. In the list of available files, look for:
   thaluna-real-time-translator-diagnostic.zip

It’s listed alongside the main version.

Once you run the Diagnostic version:

• A visible CMD console window will open together with the app
• All startup logs and errors will be printed there

If it crashes or shows errors, please click the console window, press Ctrl+A to select everything, then Enter to copy it, and paste the output here.

This will let me see exactly what’s happening on your system.

Hi! No worries, this kind of issue is usually fixable. Once I see the logs, I’ll know exactly what’s going on.

I’ve prepared a diagnostic build of Thaluna specifically for troubleshooting this problem.

Important:
The diagnostic version launches with a visible CMD console window.
All error messages and logs will be printed directly in that console, not saved to a file.

Please download and run the Diagnostic build.

When you start the app:

• Keep the CMD window open
• If any errors appear, copy the entire console output and send it to me

Additionally, please make sure you have the required Visual C++ Redistributable installed (needed by PaddleOCR):

• VC++ 2015–2022 (x64):
  https://aka.ms/vs/17/release/vc_redist.x64.exe

After installing it, restart your PC and try the diagnostic build again.

If the issue persists, please send:

• The full CMD console output
• A short description of what happens (for example: crash on startup, error message, missing files, etc.)

This thread is a hub for model recommendations and real-world feedback from both the development side and the community.

If you're unsure which model to pick for your setup, start here:

🔹 Developer's Top Pick

• gemma3:4b – The best overall balance for most users. It’s fast, smart, and highly recommended as your baseline.

🔹 Share Your Results

Hardware and use cases vary wildly, so there is no single "perfect" model. Your feedback helps everyone!

When replying, please include:

• Model Name: (e.g., gemma3:4b, llama3:8b)
• Target Language(s): (Which languages are you testing?)
• Your Hardware: (CPU, GPU, and VRAM/RAM)
• Use Case: (VN, anime, gaming, etc.)
• Priority: (Are you chasing raw speed or maximum quality?)

⚠️ Disclaimer: Thaluna is an independent project and is not affiliated with, endorsed by, or partnered with Ollama. Furthermore, Thaluna does not bundle or distribute any Ollama models within the application. Users are responsible for their own local model management.

Have a question about Quick Translator? Ask away! This section is for anything related to everyday usage.

Post here if you are unsure about:

• Real-time Streaming: How the "writing" effect works or how to adjust it.
• Language Support: How to add or switch between target languages.
• Model Manager: Questions about downloading or deleting AI models.
• Performance: How to make the translation faster on your hardware.

⚠️ IMPORTANT:

• ❌ Found a Bug? Don't post it here! Use the [Bugs / Issues] section so I can track it properly.
• 📥 Check Ollama: Most "it's not working" questions are solved by starting Ollama first!

📝 Bug Report Template :

Please fill out the information below. It helps me find the bug faster

1. What is the problem? (Describe what happened. Did the text disappear? Did the AI stop writing mid-sentence?)

2. When did it happen?

• [ ] Right after launching the app
• [ ] While downloading a model in the Model Manager
• [ ] During real-time translation (Streaming)
• [ ] When clicking "Copy" or changing languages

3. Model & Ollama Status

• Model used: (e.g., Llama 3, Mistral, etc.)
• Is Ollama running? (Check your system tray icon 📥)
• Connection check: Does http://localhost:11434"><a href="http://localhost:11434">http://localhost:11434</a> show "Ollama is running"?

4. System Specs

• Windows Version: (e.g., Win 10/11)
• RAM & GPU/CPU: (Important for AI performance!)

📂 How to find your Log Files

If something went wrong, the log file has the answer. Please attach it!

1. Press Win + R on your keyboard.
2. Paste this path and hit Enter: %LOCALAPPDATA%\ThalunaTextTranslator\logs
3. Find the file with today's date (e.g., app_20260201.log).
4. Attach the file to your post or paste the error text here.

Note: Logs are automatically deleted after 7 days to keep your system clean. 

⚠️ Requirement

Ollama is required. Thaluna Quick Translator is a local AI tool. If Ollama is not running in the background, translation will not work!

📝 Bug Report Template :

Please fill out this form to help me track down the issue.

1. What happened? (Describe what you were trying to do. Did it crash during a specific file?)

2. Model & Ollama Status

• AI Model used: (e.g., translategemma, llama3)
• Is Ollama running? (Check your system tray 📥)
• Connection check: Does <a href="http://localhost:11434">http://localhost:11434</a> say "Ollama is running"?

3. System Info

• Windows Version: (e.g., Win 10/11)
• Hardware: (e.g., 16GB RAM / RTX 3060)

📂 How to attach Log Files

If the translation failed or the app crashed, I need your logs!

1. Press Win + R on your keyboard.
2. Paste this path and hit Enter: %LOCALAPPDATA%\ThalunaSRTTranslator\logs
3. Find the file named app_YYYYMMDD.log with today's date.
4. Attach the file or copy the error text here.

Logs are auto-deleted after 7 days.

⚠️ Requirement

Ollama is required. This app translates subtitles locally using AI. If Ollama is not active, the "Translate" button won't do anything!

Use this section for general questions about using Thaluna SRT Translator.

Post here if you are unsure about:

• How SRT translation works

• Supported subtitle formats (.srt)

• Choosing the right translation model

• Translation speed and progress behavior

• Whether a result or behavior is expected

IMPORTANT:

• ❌ Do NOT report bugs here

• 🐞 Use the Bugs / Issues section for technical problems

• Please check existing threads before posting

This is a free local tool, focused on simplicity and offline use.

Hi! No worries about your English

Yes, this is possible. In many cases this happens when the game is running in exclusive fullscreen.
Thaluna works best when the game is in windowed or borderless window mode.

You can quickly check this:

• Start the game

• While the game is running, click on the Thaluna window so it has focus

• Try translating text again

If translation starts working after Thaluna gets focus, it means the game is blocking screen capture in fullscreen mode. Switching the game to windowed mode should fix it.

If Thaluna works in other games but not in this one, it’s very likely a fullscreen or emulator-specific limitation.

Regarding NDS / emulator games:
It can work, but results depend on:

• emulator rendering method

• fullscreen vs windowed mode

• font and visual style used by the game

If the translation works but the text recognition is inaccurate, you can try enabling Pixel OCR / Artistic OCR mode in settings. Some games use stylized fonts that are harder to detect in standard OCR mode.

Give it a shot and let me know if it works!

# 💰 Refund Policy & Eligibility

Thaluna is a one-time purchase software. Before purchasing, please review our refund guidelines to ensure a clear understanding of our support and return process.

### 🔍 Try Before You Buy (Free Demo)

A **free demo version** is available and **strongly recommended** before purchasing the full version. The demo allows you to verify:

* Basic OCR functionality in your specific environment.

* Real-time capture behavior and overlay performance.

* General compatibility with your system and display configuration.

*By purchasing the full version, it is assumed that the demo was either tested or intentionally skipped.*

---

### ✅ Refund Eligibility

Refunds are handled on a case-by-case basis. You may be eligible for a refund if:

1. The issue is **reproducible** and confirmed to be caused by a defect in Thaluna's code.

2. Necessary **diagnostic information** (logs or screenshots) was provided for troubleshooting.

3. The issue **cannot be resolved** after reasonable support attempts.

### ❌ Refunds will NOT be issued if:

* **No logs or screenshots are provided** (we cannot fix what we cannot see).

* **The issue cannot be reproduced** on other systems.

* **The problem is caused by local system configuration** (e.g., antivirus interference, third-party overlays, or missing Windows dependencies).

* **Specific OCR Mode Issues:** If one mode works (e.g., Manga OCR) while another doesn't (e.g., Paddle OCR), it confirms that Thaluna’s capture system is functional. Such issues are usually tied to local dependencies (like missing C++ Redistributables). **A refund will only be considered if you provide logs and we are unable to resolve the issue through support.**

* **The request is based on personal expectations** or comparisons to other tools.

* **The software was purchased for a short-term task** and a refund is requested shortly after.

---

### 💡 Important Technical Notes

* **OCR Nature:** Accuracy depends heavily on fonts, rendering methods, and system configuration.

* **Requirements:** For optimal performance, ensure **Windows display scaling is set to 100%** and the capture overlay is placed correctly over the text.

**If you encounter any issues, please open a support thread first.** I am committed to helping you get the software running perfectly, provided we can work together to diagnose the problem!

🛠 Help & Troubleshooting – Read before posting

⚠️ IMPORTANT: Windows Display Scaling For best OCR accuracy and real-time capture compatibility, Windows Display Scaling MUST be set to 100%.

• How: Right-click Desktop → Display Settings → Scale → 100%
• Symptoms: Scaling other than 100% causes misalignment or empty OCR results.

🧩 Startup & GPU Issues

• Missing DLL / ImportError: Install Microsoft Visual C++ Redistributable (x64).
• NVIDIA GPU: Make sure your drivers are up to date from the official NVIDIA website.
• Note: Restart your PC after installation.

If Windows prevents Thaluna from starting the first time, simply restart your computer and launch the app again.
This can occasionally happen with newly released applications that Windows has not seen before.

⚠️ OCR or Translation Is Incorrect / Broken If OCR reads text incorrectly, shows garbage, or does nothing—especially on RTX 5000, AMD, Intel GPUs, or very new hardware—follow these steps:

Step 1: Force CPU mode

1. Go to Settings → Performance → Hardware Acceleration.
2. Set OCR Device to CPU.
3. Set Translation Device to CPU.
4. Restart Thaluna and test again.

Step 2: Faster & Better Translation (Recommended) If CPU translation feels slow:

1. Install Ollama.
2. Select an Ollama model inside Thaluna.

• Note: Ollama will automatically use your GPU if available, even if the Translation Device is set to CPU.

🧪 Diagnostic Build & Bug Reports (MANDATORY) For crashes or OCR failures:

1. Use the Diagnostic build.
2. When an error occurs, click the console window.
3. Press Ctrl + A, then Enter.
4. Paste the full console output into your forum post.

• Bug reports without this output may be closed.

⏳ Performance & Initial Setup

• First Launch: Startup may take longer while models are prepared.
• Ollama Models: If you see blank bubbles, restart Thaluna.
• Hardware Note: Running OCR on CPU and translation via Ollama is the most stable setup.

📸 OCR Tips & Usage

• Pink Overlay Box: Real-time mode only translates text inside the pink box.
• Manga / Pixel OCR: Use these only for vertical or stylized text.
• Matching Languages: The OCR language must match the on-screen text.

💰 Refunds For refund eligibility, please refer to our Refund Policy Thread.

# ❓ Questions & Usage Help

Use this section for **general questions** and help with using Thaluna.

**Post here if you are unsure about:**

* OCR modes (Standard, Manga, Pixel, etc.)

* Translation models and language selection

* Lens Mode vs Real-time Mode

* Performance settings or recommended configurations

* Whether a behavior you see is "expected" or normal

**❗ IMPORTANT:**

* Do **NOT** post bug reports here. Use the Bug Reports section for that.

* Make sure you are using the **latest version** and check existing threads before posting.

🔧 Bug Reports / Technical Issues

Please create a NEW THREAD to report a bug. Do not post reports under existing threads unless you are experiencing the exact same issue. If you are unsure about settings or usage, please post in Questions & Usage Help instead.

🐞 Bug Report Template

To diagnose issues efficiently, logs and screenshots are mandatory. Reports without sufficient technical information may be closed. Please copy and fill out the template below in your new thread:

1. Thaluna version: Stable / Experimental
2. Windows version: (e.g. Windows 10, Windows 11)
3. CPU & GPU: (Your hardware models)
4. Target Context: (Language on screen / Game or App name)
5. Mode: Real-time / Lens / Manga OCR
6. OCR language & Translation: (From → To)
7. OCR device: CPU / GPU
8. Translation device: CPU / GPU
9. What happens?
10. Details: Describe exactly what you see on screen.
(Briefly describe: Nothing happens / Empty box / Error message)

🔧Required Files

To properly diagnose issues, you MUST use the Diagnostic build of Thaluna.

1️⃣ Diagnostic Console Output (REQUIRED)

• Download and run the Diagnostic build (listed alongside the main download).

• A CMD console window will open automatically.

• When the problem occurs:

  1. Click the console window

  2. Press Ctrl + A, then Enter

  3. Paste the full console output directly into your post

⚠️ This is the primary and required diagnostic information.

2️⃣ Log Files (ONLY if Diagnostic build cannot be used)

⚠️ Use this ONLY if you cannot run the Diagnostic build

• Standard users:

  %LOCALAPPDATA%\Thaluna\logs\ 
  

• Users with non-ASCII Windows usernames:

  C:\Users\Public\Thaluna\logs\ 
  

Attach the most recent .log file.

3️⃣ Screenshot (REQUIRED)

Provide a screenshot showing:

• Thaluna running

• Real-time mode enabled

• The pink OCR overlay box placed over the text you want to translate

❗ Why this is required

Without Diagnostic console output, issues cannot be reliably identified.
Reports without console output or screenshots may be closed.

Thank you for helping improve Thaluna.

Technical support and bug reports have been moved to the forum to keep everything organized and easier to troubleshoot. Bug reports require logs or screenshots to be actionable. Reports without sufficient technical information may be closed. Please choose the appropriate section below.

Thanks for the detailed testing.
Without logs or screenshots, I can’t take this any further.

So far, this is the only reported case where real-time translation shows absolutely no output.

To continue, I need both:

• Console logs from the latest Experimental version while using Standard / Real-time OCR mode
  (either a screenshot or copied text pasted here — whatever is easier)

• A screenshot showing real-time mode active, with the pink box placed over the text and nothing being translated

Without these two, I’m not able to diagnose or help with this issue.

Thanks for getting back to me. Since the basic setup looks correct, I’d like to do one final, focused check to understand exactly where the issue is happening on your system.

I’ve just pushed a new update to the Experimental version, so please make sure you update it first before testing.

Please go through the points below and send whatever information you can.

1. App state

Is the app paused?

Look at the Thaluna window, is there a pause icon visible?

Try clicking the play button or pressing the pause hotkey.

2. System information

What is your exact Windows version? (e.g. Windows 10, Windows 11)

What CPU and GPU do you have?

3. OCR & translation setup

What OCR language did you set? (Japanese, Korean, Chinese, English?)

What language are you trying to translate from and to?

Does the selected OCR language match the actual text you are trying to capture?

What translation model are you using? (e.g. Japanese → English — which one exactly?)

Which device is selected for OCR? (CPU or GPU?)

Which device is selected for translation? (CPU or GPU?)

4. Mode & capture behavior

Are you using Lens Mode or Real-time Mode? Does the issue happen in both?

In Real-time Mode, are you moving the transparent pink overlay box directly over the text you want to translate?

(The text must be inside the pink box not in the settings window or elsewhere.)

Try switching to Manga OCR in settings, does it work then?

5. Display scaling

Right-click Desktop → Display Settings → Scale

Is it 100%, 125%, or 150%?

Please try setting it to 100%, then test again.

6. What exactly happens when you try to translate?

Nothing at all?

Empty translation box?

Garbage / incorrect text?

Any error message?

7. Console output

The Experimental build shows runtime logs automatically.

Please try translating something and send a screenshot of the console window with logs.

Hey, just a heads-up: my previous reply below includes instructions for testing a newly updated experimental build.

Could you please try that version and let me know if the issue still occurs