Optional setups for RealSub. None of these are required — RealSub works out of the box with the bundled models and the default translation source. These guides are for people who want a translator that is faster, cheaper, or fully offline.
Status: planned. The articles below are not written yet. Watch this repository or check back after a release to see them appear.
Running a translation model on your own machine and pointing RealSub at it, so that no text leaves your computer at all and there is no per-request latency over the internet.
Planned scope:
llama.cpp / llama-server on Windows and picking a quantized model file;http://127.0.0.1:8178), and how to verify it is actually being used;Licensing note (important): models such as Sakura are distributed under terms that prohibit commercial use, which is why RealSub does not bundle or ship any of them. This guide will explain how to deploy a model you obtained yourself; complying with that model’s license is your responsibility.
For users in mainland China who want lower latency than the default source, or anyone who wants to plug an OpenAI-compatible endpoint into RealSub.
Planned scope:
https://open.bigmodel.cn/api/paas/v4, your key, model name glm-4-flash;RealSub does not bundle, pre-fill or proxy any third-party account. This is the same procedure summarized in FAQ Q2.
Have a setup worth writing up? Post it in Discussions — good write-ups can be adopted here with credit.