Comfy-Org/ComfyUI · error · NotImplementedError

There is no ddim discretization method called "{ddim_discr_m

Error message

There is no ddim discretization method called "{ddim_discr_method}"

What it means

Raised by make_ddim_timesteps when ddim_discr_method is neither 'uniform' nor 'quad'. This helper picks which DDPM timesteps the DDIM sampler uses; the method name comes from sampler parameters and must match exactly. Any other string raises NotImplementedError.

Source

Thrown at comfy/ldm/modules/diffusionmodules/util.py:128

        )

    elif schedule == "sqrt_linear":
        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64)
    elif schedule == "sqrt":
        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5
    else:
        raise ValueError(f"schedule '{schedule}' unknown.")
    return betas


def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True):
    if ddim_discr_method == 'uniform':
        c = num_ddpm_timesteps // num_ddim_timesteps
        ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
    elif ddim_discr_method == 'quad':
        ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int)
    else:
        raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"')

    # assert ddim_timesteps.shape[0] == num_ddim_timesteps
    # add one to get the final alpha values right (the ones from first scale to data during sampling)
    steps_out = ddim_timesteps + 1
    if verbose:
        logging.info(f'Selected timesteps for ddim sampler: {steps_out}')
    return steps_out


def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
    # select alphas for computing the variance schedule
    alphas = alphacums[ddim_timesteps]
    alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())

    # according to the formula provided in https://arxiv.org/abs/2010.02502
    sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
    if verbose:
        logging.info(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Use 'uniform' (even timestep spacing) or 'quad' (quadratic spacing)
2. If you need another discretization, add the branch to make_ddim_timesteps rather than passing an unknown name
3. Map external naming (e.g. 'leading') to 'uniform' at your adapter boundary

Example fix

# before
steps = make_ddim_timesteps('leading', 50, 1000)
# after
steps = make_ddim_timesteps('uniform', 50, 1000)

if ddim_discr_method not in ('uniform', 'quad'):
    raise ValueError(f"ddim_discr_method must be 'uniform' or 'quad', got {ddim_discr_method!r}")
steps = make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps)

def is_valid_discr_method(m: str) -> bool:
    return m in ('uniform', 'quad')

When it happens

Trigger: Calling make_ddim_timesteps('leading', ...) or constructing a DDIM sampler with a custom ddim_discr_method value like 'linspace' or 'trailing'.

Common situations: Copying sampler parameter dicts from other repos (k-diffusion naming, HuggingFace schedulers use 'leading'/'trailing'); extending sampler code and forgetting to add the branch; typos in the method string.

Related errors

• schedule '{schedule}' unknown.
• `only_cross_attention` can only be set to True if `added_kv_
• Unknown normalization type: {norm_type}
• Unknown activation type: {activation_type}
• Block with {block_type=} is not supported.

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/b8dfcf42841cdb06. Report an issue: GitHub.