invoke-ai/InvokeAI · error · ValueError

Invalid cfg_scale_end_step. Out of range: {cfg_scale_end_ste

Error message

Invalid cfg_scale_end_step. Out of range: {cfg_scale_end_step}.

What it means

Analogous to the start-step check: after resolving negative indices, cfg_scale_end_step must satisfy 0 <= end < num_steps. Otherwise it would fall outside the per-step cfg list and the invocation raises with the invalid value.

Source

Thrown at invokeai/app/invocations/flux_denoise.py:655

        elif isinstance(cfg_scale, list):
            cfg_scale_list = cfg_scale
        else:
            raise ValueError(f"Unsupported cfg_scale type: {type(cfg_scale)}")
        assert len(cfg_scale_list) == num_steps

        # Handle negative indices for cfg_scale_start_step and cfg_scale_end_step.
        start_step_index = cfg_scale_start_step
        if start_step_index < 0:
            start_step_index = num_steps + start_step_index
        end_step_index = cfg_scale_end_step
        if end_step_index < 0:
            end_step_index = num_steps + end_step_index

        # Validate the start and end step indices.
        if not (0 <= start_step_index < num_steps):
            raise ValueError(f"Invalid cfg_scale_start_step. Out of range: {cfg_scale_start_step}.")
        if not (0 <= end_step_index < num_steps):
            raise ValueError(f"Invalid cfg_scale_end_step. Out of range: {cfg_scale_end_step}.")
        if start_step_index > end_step_index:
            raise ValueError(
                f"cfg_scale_start_step ({cfg_scale_start_step}) must be before cfg_scale_end_step "
                + f"({cfg_scale_end_step})."
            )

        # Set values outside the start and end step indices to 1.0. This is equivalent to disabling cfg_scale for those
        # steps.
        clipped_cfg_scale = [1.0] * num_steps
        clipped_cfg_scale[start_step_index : end_step_index + 1] = cfg_scale_list[start_step_index : end_step_index + 1]

        return clipped_cfg_scale

    def _prep_inpaint_mask(self, context: InvocationContext, latents: torch.Tensor) -> torch.Tensor | None:
        """Prepare the inpaint mask.

        - Loads the mask
        - Resizes if necessary

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Set cfg_scale_end_step within 0 <= end < num_steps for the current run.
2. Use -1 to denote the final step instead of a hard-coded index.
3. Clamp: end_index = min(max(cfg_scale_end_step, -num_steps), num_steps - 1) before calling.

Example fix

// before
prep_cfg_scale(num_steps=4, cfg_scale_end_step=20) // out of range
// after
prep_cfg_scale(num_steps=4, cfg_scale_end_step=-1) // final step

if not (0 <= cfg_scale_end_step < num_steps or -num_steps <= cfg_scale_end_step < 0):
    raise ValueError('cfg_scale_end_step out of range')

def is_valid_end_step(step: int, num_steps: int) -> bool:
    return -num_steps <= step < num_steps

try:
    cfg_list = denoise.prep_cfg_scale(cfg_scale, num_steps, cfg_scale_start_step=s, cfg_scale_end_step=e)
except ValueError as e:
    if 'Invalid cfg_scale_end_step' in str(e):
        cfg_list = denoise.prep_cfg_scale(cfg_scale, num_steps, cfg_scale_start_step=s, cfg_scale_end_step=-1)
    else:
        raise

When it happens

Trigger: Calling prep_cfg_scale with end_step >= num_steps or < -num_steps (e.g. end_step=20 for a 4-step Schnell run); the end-step slider retaining a value from a longer run.

Common situations: Reducing step count in the UI without adjusting the end step; editing workflow JSON manually; reusing a shared CFG-schedule node across graphs with different step counts.

Related errors

• Invalid cfg_scale_start_step. Out of range: {cfg_scale_start
• cfg_scale_start_step ({cfg_scale_start_step}) must be before
• Unsupported cfg_scale type: {type(cfg_scale)}
• cfg_scale values must be finite.
• cfg_scale must be greater than 1

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/a2b29df075e7deea. Report an issue: GitHub.