invoke-ai/InvokeAI · error · ValueError

Invalid generation_devices value '{v}'. Use 'auto' or a list

Error message

Invalid generation_devices value '{v}'. Use 'auto' or a list of devices, e.g. ['cuda:0', 'cuda:1'].

What it means

validate_generation_devices accepts either the literal string 'auto' or an explicit list of device strings. Supplying a non-'auto' string is rejected outright because iterating it would match character-by-character ('c','u','d'...) and produce a confusing per-character error. Empty lists and malformed device names have their own messages (errors 586/587).

Source

Thrown at invokeai/app/services/config/config_default.py:290

        default=None, description="API key for Seedream image generation."
    )
    external_seedream_base_url: Optional[str] = Field(
        default=None, description="Base URL override for Seedream image generation."
    )

    # fmt: on

    model_config = SettingsConfigDict(env_prefix="INVOKEAI_", env_ignore_empty=True)

    @field_validator("generation_devices")
    @classmethod
    def validate_generation_devices(cls, v: Union[str, list[str]]) -> Union[str, list[str]]:
        if v == "auto":
            return v
        # A non-"auto" string would otherwise be iterated character-by-character below (rejecting
        # 'c' from "cuda:0"), producing a confusing error. Require an explicit list instead.
        if isinstance(v, str):
            raise ValueError(
                f"Invalid generation_devices value '{v}'. Use 'auto' or a list of devices, e.g. ['cuda:0', 'cuda:1']."
            )
        if len(v) == 0:
            raise ValueError("generation_devices cannot be an empty list. Use 'auto' or a list of devices.")
        pattern = re.compile(r"^(cpu|mps|xpu(:\d+)?|cuda(:\d+)?)$")
        for device in v:
            if not pattern.match(device):
                raise ValueError(
                    f"Invalid generation device '{device}'. Valid values are 'auto', 'cpu', 'mps', 'cuda', 'cuda:N', "
                    "'xpu', or 'xpu:N'."
                )
        return v

    @field_validator("base_url")
    @classmethod
    def validate_base_url(cls, v: Optional[str]) -> Optional[str]:
        """Normalize the reverse-proxy base path: ensure a single leading slash, no trailing slash.

View on GitHub (pinned to 0b6a024f2f)

Solutions

Change the value to 'auto' to let InvokeAI pick devices automatically
Use an explicit YAML list: generation_devices: ['cuda:0', 'cuda:1']
For env vars, use JSON list syntax, e.g. GENERATION_DEVICES='["cuda:0","cuda:1"]'

Example fix

// before (invokeai.yaml)
generation_devices: cuda:0
// after
generation_devices:
  - cuda:0
  - cuda:1

Defensive patterns

Strategy: validation

Validate before calling

v = cfg.generation_devices
assert v == 'auto' or isinstance(v, list), "generation_devices must be 'auto' or a list like ['cuda:0']"

Try / catch

try:
    cfg = InvokeAIAppConfig(**overrides)
except ValueError as e:
    if 'generation_devices' in str(e):
        overrides['generation_devices'] = 'auto'
        cfg = InvokeAIAppConfig(**overrides)

Prevention

Use 'auto' unless multi-device scheduling is needed
In YAML write lists with '- ' entries, not a single string
Use JSON list syntax for env vars

When it happens

Trigger: Configuring generation_devices as a plain string like 'cuda:0' or 'cuda' instead of a list, e.g. in invokeai.yaml or via the corresponding env var; quoting a list so YAML/env parsing yields a single string.

Common situations: Users following older docs that allowed a single device string; setting env GENERATION_DEVICES='cuda:0' without list syntax; YAML unquoted values interpreted as string rather than sequence.

Understand the failure class

Background: Schema validation failed / invalid input schema: payload rejected because its shape doesn't match the expected schema — this error's family across 28 libraries.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/9005367c8813149a. Report an issue: GitHub.