From 8644e494be720a2a898eb4ed771d6109fec34858 Mon Sep 17 00:00:00 2001 From: C43H66N12O12S2 <36072735+C43H66N12O12S2@users.noreply.github.com> Date: Wed, 28 Sep 2022 05:09:22 +0300 Subject: add eta to k ancestral --- modules/sd_samplers.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index 666ee1ee..17faeab1 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -39,8 +39,10 @@ samplers_for_img2img = [x for x in samplers if x.name != 'PLMS'] sampler_extra_params = { 'sample_euler':['s_churn','s_tmin','s_tmax','s_noise'], + 'sample_euler_ancestral':['eta'], 'sample_heun' :['s_churn','s_tmin','s_tmax','s_noise'], 'sample_dpm_2':['s_churn','s_tmin','s_tmax','s_noise'], + 'sample_dpm_2_ancestral':['eta'], } def setup_img2img_steps(p, steps=None): @@ -154,9 +156,9 @@ class VanillaStableDiffusionSampler: # existing code fails with cetin step counts, like 9 try: - samples_ddim, _ = self.sampler.sample(S=steps, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.ddim_eta) + samples_ddim, _ = self.sampler.sample(S=steps, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.eta) except Exception: - samples_ddim, _ = self.sampler.sample(S=steps+1, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.ddim_eta) + samples_ddim, _ = self.sampler.sample(S=steps+1, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.eta) return samples_ddim -- cgit v1.2.1 From 2ab64ec81a270c516816b5035860361ee145b9db Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Wed, 28 Sep 2022 10:49:07 +0300 Subject: emergency fix for #1199 --- modules/sd_samplers.py | 25 +++++++++++++------------ 1 file changed, 13 insertions(+), 12 deletions(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index 17faeab1..a1183997 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -3,6 +3,7 @@ import numpy as np import torch import tqdm from PIL import Image +import inspect import k_diffusion.sampling import ldm.models.diffusion.ddim @@ -38,11 +39,11 @@ samplers = [ samplers_for_img2img = [x for x in samplers if x.name != 'PLMS'] sampler_extra_params = { - 'sample_euler':['s_churn','s_tmin','s_tmax','s_noise'], - 'sample_euler_ancestral':['eta'], - 'sample_heun' :['s_churn','s_tmin','s_tmax','s_noise'], - 'sample_dpm_2':['s_churn','s_tmin','s_tmax','s_noise'], - 'sample_dpm_2_ancestral':['eta'], + 'sample_euler': ['s_churn', 's_tmin', 's_tmax', 's_noise'], + 'sample_euler_ancestral': ['eta'], + 'sample_heun': ['s_churn', 's_tmin', 's_tmax', 's_noise'], + 'sample_dpm_2': ['s_churn', 's_tmin', 's_tmax', 's_noise'], + 'sample_dpm_2_ancestral': ['eta'], } def setup_img2img_steps(p, steps=None): @@ -231,7 +232,7 @@ class KDiffusionSampler: self.model_wrap = k_diffusion.external.CompVisDenoiser(sd_model, quantize=shared.opts.enable_quantization) self.funcname = funcname self.func = getattr(k_diffusion.sampling, self.funcname) - self.extra_params = sampler_extra_params.get(funcname,[]) + self.extra_params = sampler_extra_params.get(funcname, []) self.model_wrap_cfg = CFGDenoiser(self.model_wrap) self.sampler_noises = None self.sampler_noise_index = 0 @@ -278,9 +279,9 @@ class KDiffusionSampler: k_diffusion.sampling.torch = TorchHijack(self) extra_params_kwargs = {} - for val in self.extra_params: - if hasattr(p,val): - extra_params_kwargs[val] = getattr(p,val) + for param_name in self.extra_params: + if hasattr(p, param_name) and param_name in inspect.signature(self.func).parameters: + extra_params_kwargs[param_name] = getattr(p, param_name) return self.func(self.model_wrap_cfg, xi, sigma_sched, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) @@ -300,9 +301,9 @@ class KDiffusionSampler: k_diffusion.sampling.torch = TorchHijack(self) extra_params_kwargs = {} - for val in self.extra_params: - if hasattr(p,val): - extra_params_kwargs[val] = getattr(p,val) + for param_name in self.extra_params: + if hasattr(p, param_name) and param_name in inspect.signature(self.func).parameters: + extra_params_kwargs[param_name] = getattr(p, param_name) samples = self.func(self.model_wrap_cfg, x, sigmas, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) -- cgit v1.2.1 From d64b451681bdba5453723d3fe0b0681a470d8045 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Wed, 28 Sep 2022 18:09:06 +0300 Subject: added support for automatically installing latest k-diffusion added eta parameter to parameters output for generated images split eta settings into ancestral and ddim (because they have different default values) --- modules/sd_samplers.py | 83 ++++++++++++++++++++++++++------------------------ 1 file changed, 44 insertions(+), 39 deletions(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index a1183997..3588aae6 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -40,10 +40,8 @@ samplers_for_img2img = [x for x in samplers if x.name != 'PLMS'] sampler_extra_params = { 'sample_euler': ['s_churn', 's_tmin', 's_tmax', 's_noise'], - 'sample_euler_ancestral': ['eta'], 'sample_heun': ['s_churn', 's_tmin', 's_tmax', 's_noise'], 'sample_dpm_2': ['s_churn', 's_tmin', 's_tmax', 's_noise'], - 'sample_dpm_2_ancestral': ['eta'], } def setup_img2img_steps(p, steps=None): @@ -101,6 +99,8 @@ class VanillaStableDiffusionSampler: self.init_latent = None self.sampler_noises = None self.step = 0 + self.eta = None + self.default_eta = 0.0 def number_of_needed_noises(self, p): return 0 @@ -123,20 +123,29 @@ class VanillaStableDiffusionSampler: self.step += 1 return res + def initialize(self, p): + self.eta = p.eta or opts.eta_ddim + + for fieldname in ['p_sample_ddim', 'p_sample_plms']: + if hasattr(self.sampler, fieldname): + setattr(self.sampler, fieldname, self.p_sample_ddim_hook) + + self.mask = p.mask if hasattr(p, 'mask') else None + self.nmask = p.nmask if hasattr(p, 'nmask') else None + def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning, steps=None): steps, t_enc = setup_img2img_steps(p, steps) # existing code fails with cetain step counts, like 9 try: - self.sampler.make_schedule(ddim_num_steps=steps, ddim_eta=p.ddim_eta, ddim_discretize=p.ddim_discretize, verbose=False) + self.sampler.make_schedule(ddim_num_steps=steps, ddim_eta=self.eta, ddim_discretize=p.ddim_discretize, verbose=False) except Exception: - self.sampler.make_schedule(ddim_num_steps=steps+1,ddim_eta=p.ddim_eta, ddim_discretize=p.ddim_discretize, verbose=False) + self.sampler.make_schedule(ddim_num_steps=steps+1, ddim_eta=self.eta, ddim_discretize=p.ddim_discretize, verbose=False) x1 = self.sampler.stochastic_encode(x, torch.tensor([t_enc] * int(x.shape[0])).to(shared.device), noise=noise) - self.sampler.p_sample_ddim = self.p_sample_ddim_hook - self.mask = p.mask if hasattr(p, 'mask') else None - self.nmask = p.nmask if hasattr(p, 'nmask') else None + self.initialize(p) + self.init_latent = x self.step = 0 @@ -145,11 +154,8 @@ class VanillaStableDiffusionSampler: return samples def sample(self, p, x, conditioning, unconditional_conditioning, steps=None): - for fieldname in ['p_sample_ddim', 'p_sample_plms']: - if hasattr(self.sampler, fieldname): - setattr(self.sampler, fieldname, self.p_sample_ddim_hook) - self.mask = None - self.nmask = None + self.initialize(p) + self.init_latent = None self.step = 0 @@ -157,9 +163,9 @@ class VanillaStableDiffusionSampler: # existing code fails with cetin step counts, like 9 try: - samples_ddim, _ = self.sampler.sample(S=steps, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.eta) + samples_ddim, _ = self.sampler.sample(S=steps, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=self.eta) except Exception: - samples_ddim, _ = self.sampler.sample(S=steps+1, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=p.eta) + samples_ddim, _ = self.sampler.sample(S=steps+1, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=self.eta) return samples_ddim @@ -237,6 +243,8 @@ class KDiffusionSampler: self.sampler_noises = None self.sampler_noise_index = 0 self.stop_at = None + self.eta = None + self.default_eta = 1.0 def callback_state(self, d): store_latent(d["denoised"]) @@ -255,22 +263,12 @@ class KDiffusionSampler: self.sampler_noise_index += 1 return res - def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning, steps=None): - steps, t_enc = setup_img2img_steps(p, steps) - - sigmas = self.model_wrap.get_sigmas(steps) - - noise = noise * sigmas[steps - t_enc - 1] - - xi = x + noise - - sigma_sched = sigmas[steps - t_enc - 1:] - + def initialize(self, p): self.model_wrap_cfg.mask = p.mask if hasattr(p, 'mask') else None self.model_wrap_cfg.nmask = p.nmask if hasattr(p, 'nmask') else None - self.model_wrap_cfg.init_latent = x self.model_wrap.step = 0 self.sampler_noise_index = 0 + self.eta = p.eta or opts.eta_ancestral if hasattr(k_diffusion.sampling, 'trange'): k_diffusion.sampling.trange = lambda *args, **kwargs: extended_trange(self, *args, **kwargs) @@ -283,6 +281,25 @@ class KDiffusionSampler: if hasattr(p, param_name) and param_name in inspect.signature(self.func).parameters: extra_params_kwargs[param_name] = getattr(p, param_name) + if 'eta' in inspect.signature(self.func).parameters: + extra_params_kwargs['eta'] = self.eta + + return extra_params_kwargs + + def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning, steps=None): + steps, t_enc = setup_img2img_steps(p, steps) + + sigmas = self.model_wrap.get_sigmas(steps) + + noise = noise * sigmas[steps - t_enc - 1] + xi = x + noise + + extra_params_kwargs = self.initialize(p) + + sigma_sched = sigmas[steps - t_enc - 1:] + + self.model_wrap_cfg.init_latent = x + return self.func(self.model_wrap_cfg, xi, sigma_sched, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) def sample(self, p, x, conditioning, unconditional_conditioning, steps=None): @@ -291,19 +308,7 @@ class KDiffusionSampler: sigmas = self.model_wrap.get_sigmas(steps) x = x * sigmas[0] - self.model_wrap_cfg.step = 0 - self.sampler_noise_index = 0 - - if hasattr(k_diffusion.sampling, 'trange'): - k_diffusion.sampling.trange = lambda *args, **kwargs: extended_trange(self, *args, **kwargs) - - if self.sampler_noises is not None: - k_diffusion.sampling.torch = TorchHijack(self) - - extra_params_kwargs = {} - for param_name in self.extra_params: - if hasattr(p, param_name) and param_name in inspect.signature(self.func).parameters: - extra_params_kwargs[param_name] = getattr(p, param_name) + extra_params_kwargs = self.initialize(p) samples = self.func(self.model_wrap_cfg, x, sigmas, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) -- cgit v1.2.1 From d62954c2bc149053f9f51dfe95751b9e0ea29f03 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Wed, 28 Sep 2022 22:30:52 +0300 Subject: fix broken DDIM with img2img --- modules/sd_samplers.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index 3588aae6..fc0c94b4 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -136,6 +136,8 @@ class VanillaStableDiffusionSampler: def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning, steps=None): steps, t_enc = setup_img2img_steps(p, steps) + self.initialize(p) + # existing code fails with cetain step counts, like 9 try: self.sampler.make_schedule(ddim_num_steps=steps, ddim_eta=self.eta, ddim_discretize=p.ddim_discretize, verbose=False) @@ -144,8 +146,6 @@ class VanillaStableDiffusionSampler: x1 = self.sampler.stochastic_encode(x, torch.tensor([t_enc] * int(x.shape[0])).to(shared.device), noise=noise) - self.initialize(p) - self.init_latent = x self.step = 0 -- cgit v1.2.1 From b05355770ce3d3512f23a3fe9681229598a0bbcf Mon Sep 17 00:00:00 2001 From: C43H66N12O12S2 <36072735+C43H66N12O12S2@users.noreply.github.com> Date: Thu, 29 Sep 2022 10:15:38 +0300 Subject: add new samplers --- modules/sd_samplers.py | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index fc0c94b4..2fb57b7d 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -23,6 +23,8 @@ samplers_k_diffusion = [ ('Heun', 'sample_heun', ['k_heun']), ('DPM2', 'sample_dpm_2', ['k_dpm_2']), ('DPM2 a', 'sample_dpm_2_ancestral', ['k_dpm_2_a']), + ('DPM fast', 'sample_dpm_fast', ['k_dpm_fast']), + ('DPM adaptive', 'sample_dpm_adaptive', ['k_dpm_ad']), ] samplers_data_k_diffusion = [ @@ -37,6 +39,8 @@ samplers = [ SamplerData('PLMS', lambda model: VanillaStableDiffusionSampler(ldm.models.diffusion.plms.PLMSSampler, model), []), ] samplers_for_img2img = [x for x in samplers if x.name != 'PLMS'] +samplers_for_img2img.remove(samplers_for_img2img[6]) +samplers_for_img2img.remove(samplers_for_img2img[6]) sampler_extra_params = { 'sample_euler': ['s_churn', 's_tmin', 's_tmax', 's_noise'], @@ -309,7 +313,12 @@ class KDiffusionSampler: x = x * sigmas[0] extra_params_kwargs = self.initialize(p) - + if 'sigma_min' in inspect.signature(self.func).parameters: + if 'n' in inspect.signature(self.func).parameters: + samples = self.func(self.model_wrap_cfg, x, sigma_min=self.model_wrap.sigmas[0].item(), sigma_max=self.model_wrap.sigmas[-1].item(), n=steps, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) + return samples + samples = self.func(self.model_wrap_cfg, x, sigma_min=self.model_wrap.sigmas[0].item(), sigma_max=self.model_wrap.sigmas[-1].item(), extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) + return samples samples = self.func(self.model_wrap_cfg, x, sigmas, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) return samples -- cgit v1.2.1 From 965dcf446991eca02074a9666048f50540261ba5 Mon Sep 17 00:00:00 2001 From: C43H66N12O12S2 <36072735+C43H66N12O12S2@users.noreply.github.com> Date: Thu, 29 Sep 2022 13:30:33 +0300 Subject: improve code quality --- modules/sd_samplers.py | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) (limited to 'modules/sd_samplers.py') diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index 2fb57b7d..5642b870 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -38,9 +38,7 @@ samplers = [ SamplerData('DDIM', lambda model: VanillaStableDiffusionSampler(ldm.models.diffusion.ddim.DDIMSampler, model), []), SamplerData('PLMS', lambda model: VanillaStableDiffusionSampler(ldm.models.diffusion.plms.PLMSSampler, model), []), ] -samplers_for_img2img = [x for x in samplers if x.name != 'PLMS'] -samplers_for_img2img.remove(samplers_for_img2img[6]) -samplers_for_img2img.remove(samplers_for_img2img[6]) +samplers_for_img2img = [x for x in samplers if x.name not in ['PLMS', 'DPM fast', 'DPM adaptive']] sampler_extra_params = { 'sample_euler': ['s_churn', 's_tmin', 's_tmax', 's_noise'], @@ -314,12 +312,12 @@ class KDiffusionSampler: extra_params_kwargs = self.initialize(p) if 'sigma_min' in inspect.signature(self.func).parameters: + extra_params_kwargs['sigma_min'] = self.model_wrap.sigmas[0].item() + extra_params_kwargs['sigma_max'] = self.model_wrap.sigmas[-1].item() if 'n' in inspect.signature(self.func).parameters: - samples = self.func(self.model_wrap_cfg, x, sigma_min=self.model_wrap.sigmas[0].item(), sigma_max=self.model_wrap.sigmas[-1].item(), n=steps, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) - return samples - samples = self.func(self.model_wrap_cfg, x, sigma_min=self.model_wrap.sigmas[0].item(), sigma_max=self.model_wrap.sigmas[-1].item(), extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) - return samples - samples = self.func(self.model_wrap_cfg, x, sigmas, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) - + extra_params_kwargs['n'] = steps + else: + extra_params_kwargs['sigmas'] = sigmas + samples = self.func(self.model_wrap_cfg, x, extra_args={'cond': conditioning, 'uncond': unconditional_conditioning, 'cond_scale': p.cfg_scale}, disable=False, callback=self.callback_state, **extra_params_kwargs) return samples -- cgit v1.2.1