The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Add the denoising step to several samplers (#8780)
chaObservf41f323c52e3
diff --git a/comfy/k_diffusion/sampling.py b/comfy/k_diffusion/sampling.pyindex e231d6a3d..34218337a 100644--- a/comfy/k_diffusion/sampling.py+++ b/comfy/k_diffusion/sampling.py@@ -412,9 +412,13 @@ def sample_lms(model, x, sigmas, extra_args=None, callback=None, disable=None, ods.pop(0)if callback is not None:callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})- cur_order = min(i + 1, order)- coeffs = [linear_multistep_coeff(cur_order, sigmas_cpu, i, j) for j in range(cur_order)]- x = x + sum(coeff * d for coeff, d in zip(coeffs, reversed(ds)))+ if sigmas[i + 1] == 0:+ # Denoising step+ x = denoised+ else:+ cur_order = min(i + 1, order)+ coeffs = [linear_multistep_coeff(cur_order, sigmas_cpu, i, j) for j in range(cur_order)]+ x = x + sum(coeff * d for coeff, d in zip(coeffs, reversed(ds)))return x@@ -1067,7 +1071,9 @@ def sample_ipndm(model, x, sigmas, extra_args=None, callback=None, disable=None,d_cur = (x_cur - denoised) / t_curorder = min(max_order, i+1)- if order == 1: # First Euler step.+ if t_next == 0: # Denoising step+ x_next = denoised+ elif order == 1: # First Euler step.x_next = x_cur + (t_next - t_cur) * d_curelif order == 2: # Use one history point.x_next = x_cur + (t_next - t_cur) * (3 * d_cur - buffer_model[-1]) / 2@@ -1085,6 +1091,7 @@ def sample_ipndm(model, x, sigmas, extra_args=None, callback=None, disable=None,return x_next+#From https://github.com/zju-pi/diff-sampler/blob/main/diff-solvers-main/solvers.py#under Apache 2 licensedef sample_ipndm_v(model, x, sigmas, extra_args=None, callback=None, disable=None, max_order=4):@@ -1108,7 +1115,9 @@ def sample_ipndm_v(model, x, sigmas, extra_args=None, callback=None, disable=Nond_cur = (x_cur - denoised) / t_curorder = min(max_order, i+1)- if order == 1: # First Euler step.+ if t_next == 0: # Denoising step+ x_next = denoised+ elif order == 1: # First Euler step.x_next = x_cur + (t_next - t_cur) * d_curelif order == 2: # Use one history point.h_n = (t_next - t_cur)@@ -1148,6 +1157,7 @@ def sample_ipndm_v(model, x, sigmas, extra_args=None, callback=None, disable=Nonreturn x_next+#From https://github.com/zju-pi/diff-sampler/blob/main/diff-solvers-main/solvers.py#under Apache 2 license@torch.no_grad()@@ -1198,6 +1208,7 @@ def sample_deis(model, x, sigmas, extra_args=None, callback=None, disable=None,return x_next+@torch.no_grad()def sample_euler_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None):extra_args = {} if extra_args is None else extra_args@@ -1404,6 +1415,7 @@ def sample_res_multistep_ancestral(model, x, sigmas, extra_args=None, callback=Ndef sample_res_multistep_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_noise=s_noise, noise_sampler=noise_sampler, eta=eta, cfg_pp=True)+@torch.no_grad()def sample_gradient_estimation(model, x, sigmas, extra_args=None, callback=None, disable=None, ge_gamma=2., cfg_pp=False):"""Gradient-estimation sampler. Paper: https://openreview.net/pdf?id=o2ND9v0CeK"""@@ -1430,19 +1442,19 @@ def sample_gradient_estimation(model, x, sigmas, extra_args=None, callback=None,if callback is not None:callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})dt = sigmas[i + 1] - sigmas[i]- if i == 0:+ if sigmas[i + 1] == 0:+ # Denoising step+ x = denoised+ else:# Euler methodif cfg_pp:x = denoised + d * sigmas[i + 1]else:x = x + d * dt- else:- # Gradient estimation- if cfg_pp:++ if i >= 1:+ # Gradient estimationd_bar = (ge_gamma - 1) * (d - old_d)- x = denoised + d * sigmas[i + 1] + d_bar * dt- else:- d_bar = ge_gamma * d + (1 - ge_gamma) * old_dx = x + d_bar * dtold_d = dreturn x