What causes "Invalid CFG scale type: ${type(self.cfg_scale)}" in invoke-ai/InvokeAI?

Set cfg_scale to a float (e.g. 7.5) to use a constant guidance scale.

invoke-ai/InvokeAI · error · ValueError

Invalid CFG scale type: ${type(self.cfg_scale)}

Error message

Invalid CFG scale type: ${type(self.cfg_scale)}

What it means

_prepare_cfg_scale only accepts cfg_scale as a float (broadcast to all timesteps) or a list matching the number of timesteps. Any other type (None, str, tensor, wrong-length list handled separately) fails this check.

Source

Thrown at invokeai/app/invocations/cogview4_denoise.py:171

        ).to(device=device, dtype=dtype)

    def _prepare_cfg_scale(self, num_timesteps: int) -> list[float]:
        """Prepare the CFG scale list.

        Args:
            num_timesteps (int): The number of timesteps in the scheduler. Could be different from num_steps depending
            on the scheduler used (e.g. higher order schedulers).

        Returns:
            list[float]: _description_
        """
        if isinstance(self.cfg_scale, float):
            cfg_scale = [self.cfg_scale] * num_timesteps
        elif isinstance(self.cfg_scale, list):
            assert len(self.cfg_scale) == num_timesteps
            cfg_scale = self.cfg_scale
        else:
            raise ValueError(f"Invalid CFG scale type: {type(self.cfg_scale)}")

        return cfg_scale

    def _convert_timesteps_to_sigmas(self, image_seq_len: int, timesteps: torch.Tensor) -> list[float]:
        # The logic to prepare the timestep / sigma schedule is based on:
        # https://github.com/huggingface/diffusers/blob/b38450d5d2e5b87d5ff7088ee5798c85587b9635/src/diffusers/pipelines/cogview4/pipeline_cogview4.py#L575-L595
        # The default FlowMatchEulerDiscreteScheduler configs are based on:
        # https://huggingface.co/THUDM/CogView4-6B/blob/fb6f57289c73ac6d139e8d81bd5a4602d1877847/scheduler/scheduler_config.json
        # This implementation differs slightly from the original for the sake of simplicity (differs in terminal value
        # handling, not quantizing timesteps to integers, etc.).

        def calculate_timestep_shift(
            image_seq_len: int, base_seq_len: int = 256, base_shift: float = 0.25, max_shift: float = 0.75
        ) -> float:
            m = (image_seq_len / base_seq_len) ** 0.5
            mu = m * max_shift + base_shift
            return mu

View on GitHub (pinned to 0b6a024f2f)

Solutions

Set cfg_scale to a float (e.g. 7.5) to use a constant guidance scale.
If per-timestep control is needed, pass a list with exactly num_timesteps entries.
Check upstream code/config that populates cfg_scale so it cannot be None or another type.

Example fix

// before
node.cfg_scale = None
// after
node.cfg_scale = 7.5  # or [7.5] * num_timesteps

Defensive patterns

Strategy: type-guard

Validate before calling

if not isinstance(node.cfg_scale, (float, list)):
    node.cfg_scale = 7.5
elif isinstance(node.cfg_scale, list) and len(node.cfg_scale) != num_timesteps:
    node.cfg_scale = node.cfg_scale[:num_timesteps] or 7.5

Type guard

def is_valid_cfg_scale(v) -> bool:
    return isinstance(v, float) or (isinstance(v, list) and all(isinstance(x, (int, float)) for x in v))

Try / catch

try:
    output = node.invoke(context)
except (ValueError, AssertionError) as e:
    if "CFG scale" in str(e):
        node.cfg_scale = 7.5
        output = node.invoke(context)
    else:
        raise

Prevention

Always initialize cfg_scale with a float default
Validate config values before assigning to the node
Never pass None or tensors into cfg_scale

When it happens

Trigger: Running the CogView4 denoise invocation with cfg_scale set to a non-float, non-list value, or a list whose length is not asserted equal to num_timesteps (that case asserts first).

Common situations: Programmatically constructing the node with cfg_scale=None from a config that failed to load; passing a list with the wrong length (hits the assert); downstream UI/schema mismatch sending unexpected types.

Understand the failure class

Background: Invalid argument type errors: "must be of type string", "expected X, got Y", and ERR_INVALID_ARG_TYPE explained — this error's family across 15 libraries.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/00f4d5e5087a9743. Report an issue: GitHub.