Local mode and API mode in LTX Desktop
LTX Desktop generates video either locally on your machine or through the LTX API, depending on your hardware. Understanding the difference helps explain costs, API key prompts, and what to expect on your setup.
How the two modes work
Local mode runs AI generation directly on your GPU. Your machine downloads and stores the model weights (around 160GB), and all generation happens on-device. There's no per-generation API cost. This mode is available on Windows and Linux with a supported NVIDIA GPU with 16GB+ VRAM.
API mode sends generation requests to the LTX cloud backend. An LTX API key is required, and video generation consumes API credits. This is the only available mode on macOS, and on Windows/Linux machines without a qualifying GPU.
Which mode does your machine use?
| Platform / hardware | Mode |
|---|---|
| Windows or Linux with NVIDIA GPU, 16GB+ VRAM | Local |
| Windows or Linux — no NVIDIA GPU, or less than 16GB VRAM | API |
| macOS (any Apple Silicon) | API |
LTX Desktop detects your hardware automatically at startup. If you only see model options labelled "(API)" in the generation panel, your machine is in API mode.
Switching between modes
On a local-capable machine (Windows or Linux with a supported NVIDIA GPU), you can choose to generate via API instead.
- Go to Settings > General > Videos Generation.
- Enable Generate With API. If you haven't added an LTX API key yet, you'll be prompted to set it up in the API Keys tab.
To switch back to local generation, return to the same setting and turn off Generate With API.
Note that on macOS or hardware without a qualifying GPU, API mode is fixed and this setting is not shown.
How text encoding works
Even in local mode, video generation requires a text encoding step. There are two ways to handle this:
- LTX API key (recommended) — text encoding runs through the LTX API. This is completely free and speeds up generation by offloading the encoding step from your GPU.
- Local Text Encoder — go to Settings > General and click Download Text Encoder. This downloads an additional model component so encoding runs on-device. This keeps everything local but requires more VRAM and disk space.
You can switch between these in Settings.
What costs credits
- Local video generation: free
- Text encoding via API: free
- Video generation via API mode: paid, uses LTX API credits
- Retake in API mode: paid, uses LTX API credits
If you're unsure whether a generation will use credits, check which model is selected in the generation panel. API models are labelled with "(API)".