Support¶
This page says which computers, browsers and versions Latentry is made and checked for, and which releases get fixes.
Supported means it is checked before a release, and a problem there is treated as a bug. Experimental means it should work, but nobody checks it on a real machine before a release. Reports are welcome, and fixes come where they can.
Releases¶
Only the latest release gets fixes, security fixes included (see SECURITY.md). To get one, update to the latest release. Compatibility says what an update may change.
Computers¶
| System | Status | Notes |
|---|---|---|
| Windows 11, x64 | Supported | CI on every pull request |
| Linux, x64 | Supported | CI on every pull request (Ubuntu) |
| macOS 14 or newer, Apple Silicon | Experimental | CI after each merge to main and once a week; not checked on a real Mac |
| Windows 10, x64 | Not checked | Microsoft no longer updates it, so it is not a safe home for a server |
| Anything else (Arm on Windows or Linux, Intel Macs, older macOS) | Not supported | The engine has no pinned install there; see below |
Latentry itself, the web UI and the gallery, needs only Node.js. On a computer the engine does not support, you can still use Latentry with backends you run yourself.
Node.js¶
Latentry needs Node.js 22.19 or newer. CI tests the 22, 24 and 26 release lines on Linux, and 22 on Windows.
- Use a release line the Node.js project still maintains. When a line reaches its end of life, a later Latentry release may stop supporting it. The changelog says so first.
- Keep Node.js up to date within your line. Some Node.js updates fix certificate trust and security problems that Latentry depends on. The changelog points out the ones that matter.
GPUs for the built-in engine¶
The engine is installed from Setup, with a PyTorch build chosen for your GPU and driver. Its Python packages are pinned for each of the supported systems above.
| GPU | Status | Notes |
|---|---|---|
| NVIDIA, driver 580 or newer | Supported | Gets current PyTorch (CUDA 13.0) |
| NVIDIA, driver 560–579 | Supported | Current PyTorch for CUDA 12.6 |
| NVIDIA RTX 50 series, driver 570–579 | Works, not recommended | CUDA 12.8, which only an older PyTorch with known vulnerabilities supports. Setup asks for a driver update |
| NVIDIA, driver older than 560 | Works, not recommended | CUDA 11.8, the same. Setup asks for a driver update |
| NVIDIA RTX 50 series, driver older than 570 | Not supported | Setup asks for a driver update first |
| Apple Silicon (Metal) | Experimental | As for macOS above |
| AMD or Intel GPUs | Not supported | The engine runs on the CPU there. Use a backend that supports the GPU instead |
| No GPU | Works, slowly | The CPU build of PyTorch |
How much the engine draws at once depends on free video memory; see Setup and the engine.
Where Latentry has no pinned install (a system above marked not supported, or a GPU newer than the pinned PyTorch knows), the engine is installed with the newest packages instead, and Setup says so.
Browsers¶
| Browser | Status |
|---|---|
| Chrome and Edge on a computer | Supported. The browser tests run on Chromium |
| Chrome on Android | Supported, for paired phones |
| Safari (Mac, iPhone, iPad) and Firefox | Experimental |
Use an up-to-date browser.
Backends¶
Backends lists what Latentry can do with each kind. Where a page names a version that was checked, such as Forge Neo 2.30, newer versions are expected to work too. A backend's own bugs are its project's to fix, but tell us if Latentry could work around one.