The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
refactor/split various bits of code for sampling
BlenderNeko2a09e2aa2762
diff --git a/comfy/sample.py b/comfy/sample.pynew file mode 100644index 000000000..ede89890b--- /dev/null+++ b/comfy/sample.py@@ -0,0 +1,62 @@+import torch+import comfy.model_management+++def prepare_noise(latent, seed, disable_noise):+ latent_image = latent["samples"]+ if disable_noise:+ noise = torch.zeros(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")+ else:+ batch_index = 0+ if "batch_index" in latent:+ batch_index = latent["batch_index"]++ generator = torch.manual_seed(seed)+ for i in range(batch_index):+ noise = torch.randn([1] + list(latent_image.size())[1:], dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")+ noise = torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")+ return noise++def create_mask(latent, noise):+ noise_mask = None+ device = comfy.model_management.get_torch_device()+ if "noise_mask" in latent:+ noise_mask = latent['noise_mask']+ noise_mask = torch.nn.functional.interpolate(noise_mask[None,None,], size=(noise.shape[2], noise.shape[3]), mode="bilinear")+ noise_mask = noise_mask.round()+ noise_mask = torch.cat([noise_mask] * noise.shape[1], dim=1)+ noise_mask = torch.cat([noise_mask] * noise.shape[0])+ noise_mask = noise_mask.to(device)+ return noise_mask++def broadcast_cond(cond, noise):+ device = comfy.model_management.get_torch_device()+ copy = []+ for p in cond:+ t = p[0]+ if t.shape[0] < noise.shape[0]:+ t = torch.cat([t] * noise.shape[0])+ t = t.to(device)+ copy += [[t] + p[1:]]+ return copy++def load_c_nets(positive, negative):+ def get_models(cond):+ models = []+ for c in cond:+ if 'control' in c[1]:+ models += [c[1]['control']]+ if 'gligen' in c[1]:+ models += [c[1]['gligen'][1]]+ return models++ return get_models(positive) + get_models(negative)++def load_additional_models(positive, negative):+ models = load_c_nets(positive, negative)+ comfy.model_management.load_controlnet_gpu(models)+ return models++def cleanup_additional_models(models):+ for m in models:+ m.cleanup()\ No newline at end of filediff --git a/comfy/samplers.py b/comfy/samplers.pyindex 15527224e..541a8db8d 100644--- a/comfy/samplers.py+++ b/comfy/samplers.py@@ -392,6 +392,38 @@ def encode_adm(noise_augmentor, conds, batch_size, device):return conds+def calculate_sigmas(model, steps, scheduler, sampler):+ """+ Returns a tensor containing the sigmas corresponding to the given model, number of steps, scheduler type and sample technique+ """+ if not (isinstance(model, CompVisVDenoiser) or isinstance(model, k_diffusion_external.CompVisDenoiser)):+ model = CFGNoisePredictor(model)+ if model.inner_model.parameterization == "v":+ model = CompVisVDenoiser(model, quantize=True)+ else:+ model = k_diffusion_external.CompVisDenoiser(model, quantize=True)++ sigmas = None++ discard_penultimate_sigma = False+ if sampler in ['dpm_2', 'dpm_2_ancestral']:+ steps += 1+ discard_penultimate_sigma = True++ if scheduler == "karras":+ sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=float(model.sigma_min), sigma_max=float(model.sigma_max))+ elif scheduler == "normal":+ sigmas = model.get_sigmas(steps)+ elif scheduler == "simple":+ sigmas = simple_scheduler(model, steps)+ elif scheduler == "ddim_uniform":+ sigmas = ddim_scheduler(model, steps)+ else:+ print("error invalid scheduler", scheduler)++ if discard_penultimate_sigma:+ sigmas = torch.cat([sigmas[:-2], sigmas[-1:]])+ return sigmasclass KSampler:SCHEDULERS = ["karras", "normal", "simple", "ddim_uniform"]@@ -421,41 +453,19 @@ class KSampler:self.denoise = denoiseself.model_options = model_options- def _calculate_sigmas(self, steps):- sigmas = None-- discard_penultimate_sigma = False- if self.sampler in ['dpm_2', 'dpm_2_ancestral']:- steps += 1- discard_penultimate_sigma = True-- if self.scheduler == "karras":- sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=self.sigma_min, sigma_max=self.sigma_max, device=self.device)- elif self.scheduler == "normal":- sigmas = self.model_wrap.get_sigmas(steps).to(self.device)- elif self.scheduler == "simple":- sigmas = simple_scheduler(self.model_wrap, steps).to(self.device)- elif self.scheduler == "ddim_uniform":- sigmas = ddim_scheduler(self.model_wrap, steps).to(self.device)- else:- print("error invalid scheduler", self.scheduler)-- if discard_penultimate_sigma:- sigmas = torch.cat([sigmas[:-2], sigmas[-1:]])- return sigmas-def set_steps(self, steps, denoise=None):self.steps = stepsif denoise is None or denoise > 0.9999:- self.sigmas = self._calculate_sigmas(steps)+ self.sigmas = calculate_sigmas(self.model_wrap, steps, self.scheduler, self.sampler).to(self.device)else:new_steps = int(steps/denoise)- sigmas = self._calculate_sigmas(new_steps)+ sigmas = calculate_sigmas(self.model_wrap, new_steps, self.scheduler, self.sampler).to(self.device)self.sigmas = sigmas[-(steps + 1):]- def sample(self, noise, positive, negative, cfg, latent_image=None, start_step=None, last_step=None, force_full_denoise=False, denoise_mask=None):- sigmas = self.sigmas+ def sample(self, noise, positive, negative, cfg, latent_image=None, start_step=None, last_step=None, force_full_denoise=False, denoise_mask=None, sigmas=None):+ if sigmas is None:+ sigmas = self.sigmassigma_min = self.sigma_minif last_step is not None and last_step < (len(sigmas) - 1):diff --git a/nodes.py b/nodes.pyindex 48c3ee9c3..601661864 100644--- a/nodes.py+++ b/nodes.py@@ -16,6 +16,7 @@ sys.path.insert(0, os.path.join(os.path.dirname(os.path.realpath(__file__)), "coimport comfy.diffusers_convertimport comfy.samplers+import comfy.sampleimport comfy.sdimport comfy.utils@@ -739,31 +740,12 @@ class SetLatentNoiseMask:s["noise_mask"] = maskreturn (s,)-def common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent, denoise=1.0, disable_noise=False, start_step=None, last_step=None, force_full_denoise=False):- latent_image = latent["samples"]- noise_mask = Nonedevice = comfy.model_management.get_torch_device()+ latent_image = latent["samples"]- if disable_noise:- noise = torch.zeros(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")- else:- batch_index = 0- if "batch_index" in latent:- batch_index = latent["batch_index"]-- generator = torch.manual_seed(seed)- for i in range(batch_index):- noise = torch.randn([1] + list(latent_image.size())[1:], dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")- noise = torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")-- if "noise_mask" in latent:- noise_mask = latent['noise_mask']- noise_mask = torch.nn.functional.interpolate(noise_mask[None,None,], size=(noise.shape[2], noise.shape[3]), mode="bilinear")- noise_mask = noise_mask.round()- noise_mask = torch.cat([noise_mask] * noise.shape[1], dim=1)- noise_mask = torch.cat([noise_mask] * noise.shape[0])- noise_mask = noise_mask.to(device)+ noise = comfy.sample.prepare_noise(latent, seed, disable_noise)+ noise_mask = comfy.sample.create_mask(latent, noise)real_model = Nonecomfy.model_management.load_model_gpu(model)@@ -772,34 +754,10 @@ def common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive,noise = noise.to(device)latent_image = latent_image.to(device)- positive_copy = []- negative_copy = []-- control_nets = []- def get_models(cond):- models = []- for c in cond:- if 'control' in c[1]:- models += [c[1]['control']]- if 'gligen' in c[1]:- models += [c[1]['gligen'][1]]- return models-- for p in positive:- t = p[0]- if t.shape[0] < noise.shape[0]:- t = torch.cat([t] * noise.shape[0])- t = t.to(device)- positive_copy += [[t] + p[1:]]- for n in negative:- t = n[0]- if t.shape[0] < noise.shape[0]:- t = torch.cat([t] * noise.shape[0])- t = t.to(device)- negative_copy += [[t] + n[1:]]-- models = get_models(positive) + get_models(negative)- comfy.model_management.load_controlnet_gpu(models)+ positive_copy = comfy.sample.broadcast_cond(positive, noise)+ negative_copy = comfy.sample.broadcast_cond(negative, noise)++ models = comfy.sample.load_additional_models(positive, negative)if sampler_name in comfy.samplers.KSampler.SAMPLERS:sampler = comfy.samplers.KSampler(real_model, steps=steps, device=device, sampler=sampler_name, scheduler=scheduler, denoise=denoise, model_options=model.model_options)@@ -809,8 +767,8 @@ def common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive,samples = sampler.sample(noise, positive_copy, negative_copy, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask)samples = samples.cpu()- for m in models:- m.cleanup()++ comfy.sample.cleanup_additional_models(models)out = latent.copy()out["samples"] = samples