Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Make ancestral samplers more deterministic.

comfyanonymous916d1e14a93e
diff --git a/comfy/k_diffusion/sampling.py b/comfy/k_diffusion/sampling.py
index 0f7cc4ca9..d37d7dd95 100644
--- a/comfy/k_diffusion/sampling.py
+++ b/comfy/k_diffusion/sampling.py
@@ -70,8 +70,14 @@ def get_ancestral_step(sigma_from, sigma_to, eta=1.):
return sigma_down, sigma_up
-def default_noise_sampler(x):
- return lambda sigma, sigma_next: torch.randn_like(x)
+def default_noise_sampler(x, seed=None):
+ if seed is not None:
+ generator = torch.Generator(device=x.device)
+ generator.manual_seed(seed)
+ else:
+ generator = None
+
+ return lambda sigma, sigma_next: torch.randn(x.size(), dtype=x.dtype, layout=x.layout, device=x.device, generator=generator)
class BatchedBrownianTree:
@@ -168,7 +174,8 @@ def sample_euler_ancestral(model, x, sigmas, extra_args=None, callback=None, dis
return sample_euler_ancestral_RF(model, x, sigmas, extra_args, callback, disable, eta, s_noise, noise_sampler)
"""Ancestral sampling with Euler method steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
for i in trange(len(sigmas) - 1, disable=disable):
denoised = model(x, sigmas[i] * s_in, **extra_args)
@@ -189,7 +196,8 @@ def sample_euler_ancestral(model, x, sigmas, extra_args=None, callback=None, dis
def sample_euler_ancestral_RF(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1.0, s_noise=1., noise_sampler=None):
"""Ancestral sampling with Euler method steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
for i in trange(len(sigmas) - 1, disable=disable):
denoised = model(x, sigmas[i] * s_in, **extra_args)
@@ -290,7 +298,8 @@ def sample_dpm_2_ancestral(model, x, sigmas, extra_args=None, callback=None, dis
"""Ancestral sampling with DPM-Solver second-order steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
for i in trange(len(sigmas) - 1, disable=disable):
denoised = model(x, sigmas[i] * s_in, **extra_args)
@@ -318,7 +327,8 @@ def sample_dpm_2_ancestral(model, x, sigmas, extra_args=None, callback=None, dis
def sample_dpm_2_ancestral_RF(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
"""Ancestral sampling with DPM-Solver second-order steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
for i in trange(len(sigmas) - 1, disable=disable):
denoised = model(x, sigmas[i] * s_in, **extra_args)
@@ -591,7 +601,8 @@ def sample_dpmpp_2s_ancestral(model, x, sigmas, extra_args=None, callback=None,
"""Ancestral sampling with DPM-Solver++(2S) second-order steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
sigma_fn = lambda t: t.neg().exp()
t_fn = lambda sigma: sigma.log().neg()
@@ -625,7 +636,8 @@ def sample_dpmpp_2s_ancestral(model, x, sigmas, extra_args=None, callback=None,
def sample_dpmpp_2s_ancestral_RF(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
"""Ancestral sampling with DPM-Solver++(2S) second-order steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
s_in = x.new_ones([x.shape[0]])
sigma_fn = lambda lbda: (lbda.exp() + 1) ** -1
lambda_fn = lambda sigma: ((1-sigma)/sigma).log()
@@ -1153,7 +1165,8 @@ def sample_euler_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disabl
def sample_euler_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
"""Ancestral sampling with Euler method steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
temp = [0]
def post_cfg_function(args):
@@ -1179,7 +1192,8 @@ def sample_euler_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=No
def sample_dpmpp_2s_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
"""Ancestral sampling with DPM-Solver++(2S) second-order steps."""
extra_args = {} if extra_args is None else extra_args
- noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+ seed = extra_args.get("seed", None)
+ noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
temp = [0]
def post_cfg_function(args):