[Tools] Hoshi2Star ★ — Free offline CAT editor for RPG Maker MV/MZ fan translation (Ollama, no API key, Linux/Windows)

[Blackat]

Hi Anime-Sharing!

I built a free desktop tool for translating Japanese RPG Maker
MV/MZ games and wanted to share it here.

━━━ WHAT IS IT ━━━

Hoshi2Star (星 → ★) is a CAT editor — it lets you translate
Japanese RPG Maker games using a local AI model, then review
and correct every segment before applying the translation to
the game.

100% offline. No API key. No subscription. No cloud.
Your text never leaves your machine.

━━━ WHAT IT DOES ━━━

Opens RPG Maker MV/MZ games and extracts all text
Pre-translates using Ollama (local AI, runs on your machine)
CAT editor — review and edit every segment before export
Glossary auto-extraction — detects character names, items,
skills and uses them in every translation prompt
Translation Memory — reuses previous translations
automatically (fuzzy matching at 80% similarity)
QA engine — checks placeholders, line length, encoding
Translate All — translates entire project with configurable
cooldown between files
Export QA report as standalone HTML
Project list with progress tracking

━━━ RECOMMENDED MODEL ━━━

ollama pull qwen3:4b-instruct-2507-q8_0

Pick based on your VRAM:
| Model | VRAM |
|-------------------------------|-------|
| qwen3:1.7b-instruct | ~2 GB |
| qwen3:4b-instruct-2507-q8_0 | ~5 GB | ← recommended
| qwen3:8b-instruct | ~8 GB |
| qwen3:14b-instruct | ~12 GB|

━━━ REQUIREMENTS ━━━

- Ollama installed: https://ollama.ai
- Linux or Windows
- No account, no API key, no subscription

━━━ SUPPORTED ENGINES ━━━

RPG Maker MV
RPG Maker MZ
Wolf RPG (v0.4.0 — in development)
RPG Maker VX Ace (planned)

━━━ DOWNLOAD ━━━

GitHub: https://github.com/KATBlackCoder/Hoshi2Star
Latest release: https://github.com/KATBlackCoder/Hoshi2Star/releases/latest

━━━ NOTES ━━━

- MIT license, fully open source
- No telemetry, everything stays local
- Interface in English and French
- Bug reports → GitHub Issues

Happy to answer questions!

 

[Yupiel]

Is there any other models that we can use that has good results with the program? I'm not sure if Qwen 3 is a good model for JP translation and I'm sure there's other better ones. The only problem is I'm assuming you have a prompt in the program and the model can "hopefully" follow said prompt.

 

[Blackat]

Great question! Yes, you can use other models with Hoshi2Star.
The app works with any model available in Ollama. But in my search I know only Qwen 3 because I didn't find better and he do the job great for now.
Let me know if you try other models and get good results —always happy to update the recommendations!

 

[Yupiel]

Great question! Yes, you can use other models with Hoshi2Star.
The app works with any model available in Ollama. But in my search I know only Qwen 3 because I didn't find better and he do the job great for now.
Let me know if you try other models and get good results —always happy to update the recommendations!

Gemma 4 E4B works quite well. I already used the 24B model with LM Studio before but it was too slow for my RTX 3060ti. E4B is still pretty slow but leagues better on speed. Mine is on a 8GB VRAM gpu.

Not gonna lie, I really like how simple the program is. I have already translated a JP RPGM game and its working flawlessly so far. Translation with Qwen 3 is subpar but with Gemma 4 its leagues better and it still follows instruction.

If you want more info, I mostly got it from this reddit thread.

 

[Blackat]

Thanks for the detailed feedback, really appreciate it!

Gemma 4 E4B is a great find — I'll add it to the
recommended models list in the next update. Unfortunately
I can't test it myself since I only have a GTX 1660 Super
(6 GB VRAM), so your feedback is especially valuable!

Glad the app is working well for you.
Wolf RPG support is coming in the next version — stay tuned!