Connect providers
Kitly generates most things through cloud services called providers. To use one, you give Kitly an API key. If you'd rather not use the cloud at all, skip to local models below.
What an API key is, in plain words
An API key is a long password that a service gives you so apps can use your account on that service. You create it on the provider's website (usually under an "API keys" page), copy it once, and paste it into Kitly. Kitly then uses it whenever you generate something with that provider, and the provider bills your account there, not Kitly.
Treat a key like a password: don't share it or post it anywhere.
Adding a key
- Open Settings (the gear icon, or the burger menu's Settings...).
- In the API keys section, pick the provider's tab.
- Paste your key and click Save. Click Clear to remove one later.
Saved keys display masked, showing only the last four characters, with a storage note: "Stored securely" means it's kept in your system's credential store, while a key marked as session only clears when you quit the app.
The six providers and what each unlocks
| Provider | Key looks like | What it unlocks |
|---|---|---|
| Fal.ai | fal_... |
The bulk of Kitly's operations: image, 3D, splat, audio, music, and video generation. If you connect only one provider, make it this one. |
| OpenRouter | sk-or-... |
The Text channel's OpenRouter pipeline, plus the Assistant when set to OpenRouter. |
AIza... |
Gemini image, video, and speech operations, and the Assistant's Gemini chat and plan mode. | |
| OpenAI | sk-... |
OpenAI image operations and the Assistant. |
| Anthropic | sk-ant-... |
The Assistant's Claude chat and plan mode. |
| ElevenLabs | sk_... |
The direct ElevenLabs text-to-speech, voice changer, and Scribe transcription operations, including any voices you've cloned on your ElevenLabs account. |
One key can cover two providers
The checkbox Use Fal for OpenRouter routes OpenRouter calls through your Fal key, so you don't need a separate OpenRouter account for text generation.
Local models instead
You can run some operations on your own machine with no cloud account and no per-generation cost. During onboarding, the provider step has an I'll use local models instead button that turns this on. Later, manage it under Settings > Local Inference. If you have an NVIDIA graphics card, the same section has a one-click Download CUDA support (1.8 GB) button that speeds up several local models; see Settings.
The steps:
- Tick Enable local inference.
- Browse the downloadable model cards. Each is an ONNX model, a portable neural-network file format that runs on your hardware. Click Download (you'll accept the model's license first) and watch the progress bar.
- Each model card has a "Use for" choice per task: Cloud (Fal), Local (this device), or Prefer local (use the local model when it's there, fall back to the cloud when it isn't).
You can also register your own models with Add Model...: give it a title, pick the Task, and set YOLO preprocessing or normalization options if the model needs them.
In the operation browser, an amber chip on an operation means it runs on Fal right now but has a local model you could download instead. Clicking the chip opens Settings.
Next step
If you'll be working with video, set up FFmpeg next. Otherwise head straight to Sandbox Mode and generate your first asset.