UNKNOWN_MODEL: Unknown model — the exact model ids Genace accepts

{
  "error": {
    "type": "model_unavailable",
    "code": "UNKNOWN_MODEL",
    "message": "Unknown model: veo-3",
    "param": "model"
  }
}

The model string matched no registered provider. The message echoes what you sent, so the typo is usually visible in the error itself.

The accepted ids

Model ids are exact strings, not display names:

Modality	Model id	Display name
video	seedance-2.0-fast	Seedance 2.0 Fast
video	kling-2.0-standard	Kling 2.0 Standard
video	veo3	Veo 3
image	nano-banana	Nano Banana
image	flux-pro	Flux Pro v1.1
image	ideogram-v3	Ideogram v3

Two that catch people out:

• veo3, with no hyphen. Not veo-3.
• Version numbers use dots. seedance-2.0-fast, not seedance-2-0-fast.

Omitting the field

model is optional. Leave it out and you get the default for that endpoint:

• video → seedance-2.0-fast
• image → nano-banana

Both defaults are the cheapest option in their modality. That is deliberate — an unspecified model should not be the expensive one.

Listing them at runtime

Rather than hard-coding the table above, GET /v1/models returns the current registry. It needs authentication but costs nothing and touches no quota, which also makes it a convenient way to verify an API key — see INVALID_API_KEY.

This used to be a 500

Like INVALID_PARAM, an unknown model previously threw a plain error and surfaced as 500 internal server error, hiding the fact that the caller simply had a typo. It now returns 400 with the offending value.