Comfy-Org/ComfyUI · error · ValueError

Order {order} too high for step {i}

Error message

Order {order} too high for step {i}

What it means

Raised by linear_multistep_coeff, the coefficient helper for sample_lms (linear multistep sampler). An order-N Adams-Bashforth step at position i needs i+1 previous points; if order-1 > i (i.e. the very first steps of the schedule don't yet have enough history), the Lagrange product would index negative positions, so it refuses up front.

Source

Thrown at comfy/k_diffusion/sampling.py:408

        if sigma_down == 0:
            # Euler method
            dt = sigma_down - sigmas[i]
            x = x + d * dt
        else:
            # DPM-Solver-2
            sigma_mid = sigmas[i].log().lerp(sigma_down.log(), 0.5).exp()
            dt_1 = sigma_mid - sigmas[i]
            dt_2 = sigma_down - sigmas[i]
            x_2 = x + d * dt_1
            denoised_2 = model(x_2, sigma_mid * s_in, **extra_args)
            d_2 = to_d(x_2, sigma_mid, denoised_2)
            x = x + d_2 * dt_2
            x = (alpha_ip1/alpha_down) * x + noise_sampler(sigmas[i], sigmas[i + 1]) * s_noise * renoise_coeff
    return x

def linear_multistep_coeff(order, t, i, j):
    if order - 1 > i:
        raise ValueError(f'Order {order} too high for step {i}')
    def fn(tau):
        prod = 1.
        for k in range(order):
            if j == k:
                continue
            prod *= (tau - t[i - k]) / (t[i - j] - t[i - k])
        return prod
    return integrate.quad(fn, t[i], t[i + 1], epsrel=1e-4)[0]


@torch.no_grad()
def sample_lms(model, x, sigmas, extra_args=None, callback=None, disable=None, order=4):
    extra_args = {} if extra_args is None else extra_args
    s_in = x.new_ones([x.shape[0]])
    sigmas_cpu = sigmas.detach().cpu().numpy()
    ds = []
    for i in trange(len(sigmas) - 1, disable=disable):
        denoised = model(x, sigmas[i] * s_in, **extra_args)

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Clamp order during warmup: order = min(order, i + 1) exactly as sample_lms does
2. Lower the global order to <= number of early steps you can skip, or start integration after enough history exists
3. When calling linear_multistep_coeff directly, ensure i >= order-1

Example fix

# before
coeff = linear_multistep_coeff(4, t, i=1, j=1)  # order-1 > i
# after
order = min(4, i + 1)
coeff = linear_multistep_coeff(order, t, i=1, j=1)

order = min(order, i + 1)  # warmup clamp, mirrors sample_lms
coeff = linear_multistep_coeff(order, t, i, j)

When it happens

Trigger: Calling sample_lms (or linear_multistep_coeff directly) with order > number of warmup steps available — concretely order=4 is fine because sample_lms clamps with min(order, i+1), but direct calls to linear_multistep_coeff(order, t, i, j) with order-1 > i raise. Also custom samplers that forget the warmup clamp.

Common situations: Copy-pasting sample_lms into a custom sampler without the `order = min(order, i + 1)` warmup line; calling the coefficient helper standalone for unit tests with small i; schedules with very few steps combined with high order in modified code.

Related errors

• Unsupported noise schedule {}. The schedule needs to be 'dis
• sigma_min and sigma_max must not be 0
• solver_type must be 'heun' or 'midpoint'
• solver_type must be 'phi_1' or 'phi_2'
• ar_video sampler requires 5-D video latents [B,C,T,H,W], got

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/244d1166d26a3896. Report an issue: GitHub.