Models and Providers

A provider is the service that runs the model. A model is the specific AI system you pick in the generate bar.

Bragi does not bundle API keys. You add keys in plugin settings.

Providers you can connect

Each provider shows the key fields it needs. Some use multiple fields, such as access key and secret key.

Model types

Why a model may be hidden

A model may not appear in the generate bar because:

• Its provider is not configured.
• That model is not explicitly connected to the provider.
• The model is disabled in Settings → Models.
• You opened the wrong generation type (image vs video vs text vs audio).
• Upstream references are incompatible with every available model for that type.

Adding a provider

1. Open Settings → Bragi Canvas.
2. Under Providers, click +.
3. Enter your keys.
4. Select which models to enable for that provider.
5. Reopen the generate bar after saving.

New catalog models are not automatically enabled. Enable the model and connect it to a configured provider before it appears in the generate bar or agent model list.

Custom model IDs

Volcengine and BytePlus let you enter a custom API model ID or inference endpoint ID (ep-...) for every Seedance and Seedream model. Choose the matching model in Bragi, then enter the ID supplied by your provider. A custom ID changes where the request is sent; the selected model and provider still determine the available modes and controls.

For example, a BytePlus endpoint for Seedance 2.5 belongs under Seedance 2.5. Make sure the endpoint actually serves that model. Providers that choose an upstream model automatically for each mode keep that routing and do not offer an editable ID.

BreezeBlue voice references

BreezeBlue uses one API key and offers Breeze TTS 2 Multilingual for voice-design reference audio. Enable it explicitly after adding your key. Its Candidates control creates 1–3 audio files from a connected voice description and the current node's spoken line. See Audio for the workflow.

SVRouter image and video models

With plugin 1.43.0 or later, SVRouter supports GPT Image 2.5 and Fast H3 Omni. Enter your SVRouter gateway token, then explicitly enable the model and connect it to SVRouter. Connecting GPT Image 2.5 to another provider keeps your current provider selection until you change it.

Your gateway must have the matching model channel configured. Fast H3 Omni also requires a deployed Reactor worker; availability depends on the gateway and upstream service. See Image and Video for the controls and reference limits.

Wan Prime and Vidu

Plugin 1.45.0 adds Prime as a version of the existing Wan 3.0 model on DashScope, and Vidu Q4 Preview as a model you can enable on APIMart. See Video for region setup and reference requirements.

Kling 4.0

Kling 4.0 is available through APIMart in plugin 1.45.0 or later. Add your APIMart key, explicitly enable Kling 4.0, and select APIMart as its provider. See Video for its modes and keyframe notation. The public catalog above follows the latest published stable release.

Temporary cloud

Temporary cloud (Bragi Relay) briefly hosts reference files when a provider requires a fetchable URL. Temporary files are scheduled to expire after one day; deletion is asynchronous.

Generation outputs still save locally in _bragi/assets/. Temporary cloud is for reference transfer, not long-term storage.

Click Test under Temporary cloud to verify connectivity.

Security

• Store keys only in Bragi Canvas settings.
• Do not paste API keys into canvas nodes, prompts, or shared screenshots.
• Do not commit data.json from your vault to public repositories if it contains keys.

Related pages

• Install and First Key
• No Models Listed