stable-diffusion-webui@f1db987 (original) (raw)

`@@ -70,8 +70,13 @@ def before_sample(self, x, ts, cond, unconditional_conditioning):

70

71

`# Have to unwrap the inpainting conditioning here to perform pre-processing

72

`image_conditioning = None

73

uc_image_conditioning = None

73

74

`if isinstance(cond, dict):

74

image_conditioning = cond["c_concat"][0]

75

if self.conditioning_key == "crossattn-adm":

76

image_conditioning = cond["c_adm"]

77

uc_image_conditioning = unconditional_conditioning["c_adm"]

78

else:

79

image_conditioning = cond["c_concat"][0]

75

80

`cond = cond["c_crossattn"][0]

76

81

`unconditional_conditioning = unconditional_conditioning["c_crossattn"][0]

77

82

`@@ -98,8 +103,12 @@ def before_sample(self, x, ts, cond, unconditional_conditioning):

98

103

`# Wrap the image conditioning back up since the DDIM code can accept the dict directly.

99

104

`# Note that they need to be lists because it just concatenates them later.

100

105

`if image_conditioning is not None:

101

cond = {"c_concat": [image_conditioning], "c_crossattn": [cond]}

102

unconditional_conditioning = {"c_concat": [image_conditioning], "c_crossattn": [unconditional_conditioning]}

106

if self.conditioning_key == "crossattn-adm":

107

cond = {"c_adm": image_conditioning, "c_crossattn": [cond]}

108

unconditional_conditioning = {"c_adm": uc_image_conditioning, "c_crossattn": [unconditional_conditioning]}

109

else:

110

cond = {"c_concat": [image_conditioning], "c_crossattn": [cond]}

111

unconditional_conditioning = {"c_concat": [image_conditioning], "c_crossattn": [unconditional_conditioning]}

103

112

104

113

`return x, ts, cond, unconditional_conditioning

105

114

`@@ -176,8 +185,12 @@ def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning,

176

185

177

186

`# Wrap the conditioning models with additional image conditioning for inpainting model

178

187

`if image_conditioning is not None:

179

conditioning = {"c_concat": [image_conditioning], "c_crossattn": [conditioning]}

180

unconditional_conditioning = {"c_concat": [image_conditioning], "c_crossattn": [unconditional_conditioning]}

188

if self.conditioning_key == "crossattn-adm":

189

conditioning = {"c_adm": image_conditioning, "c_crossattn": [conditioning]}

190

unconditional_conditioning = {"c_adm": torch.zeros_like(image_conditioning), "c_crossattn": [unconditional_conditioning]}

191

else:

192

conditioning = {"c_concat": [image_conditioning], "c_crossattn": [conditioning]}

193

unconditional_conditioning = {"c_concat": [image_conditioning], "c_crossattn": [unconditional_conditioning]}

181

194

182

195

`samples = self.launch_sampling(t_enc + 1, lambda: self.sampler.decode(x1, conditioning, t_enc, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning))

183

196

`@@ -195,8 +208,12 @@ def sample(self, p, x, conditioning, unconditional_conditioning, steps=None, ima

195

208

`# Wrap the conditioning models with additional image conditioning for inpainting model

196

209

`# dummy_for_plms is needed because PLMS code checks the first item in the dict to have the right shape

197

210

`if image_conditioning is not None:

198

conditioning = {"dummy_for_plms": np.zeros((conditioning.shape[0],)), "c_crossattn": [conditioning], "c_concat": [image_conditioning]}

199

unconditional_conditioning = {"c_crossattn": [unconditional_conditioning], "c_concat": [image_conditioning]}

211

if self.conditioning_key == "crossattn-adm":

212

conditioning = {"dummy_for_plms": np.zeros((conditioning.shape[0],)), "c_crossattn": [conditioning], "c_adm": image_conditioning}

213

unconditional_conditioning = {"c_crossattn": [unconditional_conditioning], "c_adm": torch.zeros_like(image_conditioning)}

214

else:

215

conditioning = {"dummy_for_plms": np.zeros((conditioning.shape[0],)), "c_crossattn": [conditioning], "c_concat": [image_conditioning]}

216

unconditional_conditioning = {"c_crossattn": [unconditional_conditioning], "c_concat": [image_conditioning]}

200

217

201

218

`samples_ddim = self.launch_sampling(steps, lambda: 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)[0])

202

219