forked from showlab/X-Adapter
-
Notifications
You must be signed in to change notification settings - Fork 8
/
nodes.py
373 lines (315 loc) · 19.3 KB
/
nodes.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
import torch
import os
from diffusers import DPMSolverMultistepScheduler
from torch import Generator
from torchvision import transforms
from transformers import CLIPTokenizer, PretrainedConfig
from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, EulerDiscreteScheduler, ControlNetModel
from .xadapter.model.unet_adapter import UNet2DConditionModel as UNet2DConditionModel_v2
from .xadapter.model.adapter import Adapter_XL
from .pipeline.pipeline_sd_xl_adapter_controlnet_img2img import StableDiffusionXLAdapterControlnetI2IPipeline
from .pipeline.pipeline_sd_xl_adapter_controlnet import StableDiffusionXLAdapterControlnetPipeline
from omegaconf import OmegaConf
from .utils.single_file_utils import (create_scheduler_from_ldm, create_text_encoders_and_tokenizers_from_ldm, convert_ldm_vae_checkpoint,
convert_ldm_unet_checkpoint, create_text_encoder_from_ldm_clip_checkpoint, create_vae_diffusers_config,
create_diffusers_controlnet_model_from_ldm, create_unet_diffusers_config)
from safetensors import safe_open
import comfy.model_management
import comfy.utils
import folder_paths
script_directory = os.path.dirname(os.path.abspath(__file__))
class Diffusers_X_Adapter:
def __init__(self):
print("Initializing Diffusers_X_Adapter")
self.device = comfy.model_management.get_torch_device()
self.dtype = torch.float16 if comfy.model_management.should_use_fp16() and not comfy.model_management.is_device_mps(self.device) else torch.float32
self.current_1_5_checkpoint = None
self.current_lora = None
self.current_controlnet_checkpoint = None
self.original_config = OmegaConf.load(os.path.join(script_directory, f"configs/v1-inference.yaml"))
self.sdxl_original_config = OmegaConf.load(os.path.join(script_directory, f"configs/sd_xl_base.yaml"))
self.controlnet_original_config = OmegaConf.load(os.path.join(script_directory, f"configs/control_v11p_sd15.yaml"))
@classmethod
def IS_CHANGED(s):
return ""
@classmethod
def INPUT_TYPES(cls):
return {"required":
{
"sd_1_5_checkpoint": (folder_paths.get_filename_list("checkpoints"), ),
"lora_checkpoint": (folder_paths.get_filename_list("loras"), ),
"use_lora": ("BOOLEAN", {"default": False}),
"width_sd1_5": ("INT", {"default": 512, "min": 64, "max": 4096, "step": 8}),
"height_sd1_5": ("INT", {"default": 512, "min": 64, "max": 4096, "step": 8}),
"batch_size": ("INT", {"default": 1, "min": 1, "max": 100, "step": 1}),
"resolution_multiplier": ("INT", {"default": 2, "min": 2, "max": 2, "step": 1}),
"prompt_sd1_5": ("STRING", {"multiline": True, "default": "positive prompt sd1_5",}),
"sdxl_checkpoint": (folder_paths.get_filename_list("checkpoints"), ),
#"width_sdxl": ("INT", {"default": 1024, "min": 64, "max": 4096, "step": 8}),
#"height_sdxl": ("INT", {"default": 1024, "min": 64, "max": 4096, "step": 8}),
"prompt_sdxl": ("STRING", {"multiline": True, "default": "positive prompt sdxl",}),
"negative_prompt": ("STRING", {"multiline": True, "default": "negative",}),
"controlnet_name": (folder_paths.get_filename_list("controlnet"), ),
"guess_mode": ("BOOLEAN", {"default": False}),
"control_guidance_start": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"control_guidance_end": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"seed": ("INT", {"default": 123,"min": 0, "max": 0xffffffffffffffff, "step": 1}),
"steps": ("INT", {"default": 20, "min": 1, "max": 4096, "step": 1}),
"cfg": ("FLOAT", {"default": 8.0, "min": 0.1, "max": 100.0, "step": 0.1}),
"controlnet_condition_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0, "step": 0.01}),
"adapter_condition_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0, "step": 0.01}),
"adapter_guidance_start": ("FLOAT", {"default": 0.7, "min": 0.0, "max": 10.0, "step": 0.01}),
"use_xformers": ("BOOLEAN", {"default": False}),
},
"optional": {
"controlnet_image" : ("IMAGE",),
"latent_source_image" : ("IMAGE",),
},
}
RETURN_TYPES = ("IMAGE",)
FUNCTION = "load_checkpoint"
CATEGORY = "Diffusers-X-Adapter"
def load_checkpoint(self, prompt_sdxl, prompt_sd1_5, negative_prompt, use_xformers, sd_1_5_checkpoint, lora_checkpoint, use_lora, sdxl_checkpoint, resolution_multiplier,
controlnet_name, seed, steps, cfg, width_sd1_5, height_sd1_5, batch_size, #width_sdxl, height_sdxl,
adapter_condition_scale, adapter_guidance_start, controlnet_condition_scale, guess_mode, control_guidance_start, control_guidance_end, controlnet_image=None, latent_source_image=None):
if latent_source_image is not None:
latent_source_image = latent_source_image.permute(0, 3, 1, 2)
model_path_sd1_5 = folder_paths.get_full_path("checkpoints", sd_1_5_checkpoint)
lora_path = folder_paths.get_full_path("loras", lora_checkpoint)
model_path_sdxl = folder_paths.get_full_path("checkpoints", sdxl_checkpoint)
controlnet_path = folder_paths.get_full_path("controlnet", controlnet_name)
if not use_lora:
self.current_lora = None
if not hasattr(self, 'unet_sd1_5') or self.current_1_5_checkpoint != sd_1_5_checkpoint or self.current_lora != lora_checkpoint:
self.pipeline = None
self.unet_sd1_5 = None
comfy.model_management.soft_empty_cache()
print("Loading SD_1_5 checkpoint: ", sd_1_5_checkpoint)
self.current_1_5_checkpoint = sd_1_5_checkpoint
self.current_lora = lora_checkpoint
if model_path_sd1_5.endswith(".safetensors"):
state_dict_sd1_5 = {}
with safe_open(model_path_sd1_5, framework="pt", device="cpu") as f:
for key in f.keys():
state_dict_sd1_5[key] = f.get_tensor(key)
elif model_path_sd1_5.endswith(".ckpt"):
state_dict_sd1_5 = torch.load(model_path_sd1_5, map_location="cpu")
while "state_dict" in state_dict_sd1_5:
state_dict_sd1_5 = state_dict_sd1_5["state_dict"]
# 1. vae
converted_vae_config = create_vae_diffusers_config(self.original_config, image_size=512)
converted_vae = convert_ldm_vae_checkpoint(state_dict_sd1_5, converted_vae_config)
self.vae_sd1_5 = AutoencoderKL(**converted_vae_config)
self.vae_sd1_5.load_state_dict(converted_vae, strict=False)
self.vae_sd1_5.to(self.dtype)
# 2. unet
converted_unet_config = create_unet_diffusers_config(self.original_config, image_size=512)
converted_unet = convert_ldm_unet_checkpoint(state_dict_sd1_5, converted_unet_config)
self.unet_sd1_5 = UNet2DConditionModel_v2(**converted_unet_config)
self.unet_sd1_5.load_state_dict(converted_unet, strict=False)
self.unet_sd1_5.to(self.dtype)
# 3. text encoder and tokenizer
converted_text_encoder_and_tokenizer = create_text_encoders_and_tokenizers_from_ldm(self.original_config, state_dict_sd1_5)
self.tokenizer_sd1_5 = converted_text_encoder_and_tokenizer['tokenizer']
self.text_encoder_sd1_5 = converted_text_encoder_and_tokenizer['text_encoder']
self.text_encoder_sd1_5.to(self.dtype)
# 4. scheduler
self.scheduler_sd1_5 = create_scheduler_from_ldm("DPMSolverMultistepScheduler", self.original_config, state_dict_sd1_5, scheduler_type="ddim")['scheduler']
del state_dict_sd1_5, converted_unet, converted_vae, converted_text_encoder_and_tokenizer
# 5. lora
if use_lora:
print("Loading LoRA: ", lora_checkpoint)
self.lora_checkpoint = lora_checkpoint
if lora_path.endswith(".safetensors"):
state_dict_lora = {}
with safe_open(lora_path, framework="pt", device="cpu") as f:
for key in f.keys():
state_dict_lora[key] = f.get_tensor(key)
elif lora_path.endswith(".ckpt"):
state_dict_lora = torch.load(lora_path, map_location="cpu")
while "state_dict" in state_dict_lora:
state_dict_lora = state_dict_lora["state_dict"]
LORA_PREFIX_UNET = 'lora_unet'
LORA_PREFIX_TEXT_ENCODER = 'lora_te'
alpha = 1
visited = []
# directly update weight in diffusers model
for key in state_dict_lora:
# it is suggested to print out the key, it usually will be something like below
# "lora_te_text_model_encoder_layers_0_self_attn_k_proj.lora_down.weight"
# as we have set the alpha beforehand, so just skip
if '.alpha' in key or key in visited:
continue
if 'text' in key:
layer_infos = key.split('.')[0].split(LORA_PREFIX_TEXT_ENCODER + '_')[-1].split('_')
curr_layer = self.text_encoder_sd1_5
else:
layer_infos = key.split('.')[0].split(LORA_PREFIX_UNET + '_')[-1].split('_')
curr_layer = self.unet_sd1_5
# find the target layer
temp_name = layer_infos.pop(0)
while len(layer_infos) > -1:
try:
curr_layer = curr_layer.__getattr__(temp_name)
if len(layer_infos) > 0:
temp_name = layer_infos.pop(0)
elif len(layer_infos) == 0:
break
except Exception:
if len(temp_name) > 0:
temp_name += '_' + layer_infos.pop(0)
else:
temp_name = layer_infos.pop(0)
# org_forward(x) + lora_up(lora_down(x)) * multiplier
pair_keys = []
if 'lora_down' in key:
pair_keys.append(key.replace('lora_down', 'lora_up'))
pair_keys.append(key)
else:
pair_keys.append(key)
pair_keys.append(key.replace('lora_up', 'lora_down'))
# update weight
if len(state_dict_lora[pair_keys[0]].shape) == 4:
weight_up = state_dict_lora[pair_keys[0]].squeeze(3).squeeze(2).to(torch.float32)
weight_down = state_dict_lora[pair_keys[1]].squeeze(3).squeeze(2).to(torch.float32)
curr_layer.weight.data += alpha * torch.mm(weight_up, weight_down).unsqueeze(2).unsqueeze(3)
else:
weight_up = state_dict_lora[pair_keys[0]].to(torch.float32)
weight_down = state_dict_lora[pair_keys[1]].to(torch.float32)
curr_layer.weight.data += alpha * torch.mm(weight_up, weight_down)
# update visited list
for item in pair_keys:
visited.append(item)
del state_dict_lora, visited, pair_keys
else:
self.current_lora = None
# load controlnet
if controlnet_image is not None:
if not hasattr(self, 'controlnet') or self.current_controlnet_checkpoint != controlnet_name:
self.pipeline = None
print("Loading controlnet: ", controlnet_name)
self.current_controlnet_checkpoint = controlnet_name
if controlnet_path.endswith(".safetensors"):
state_dict_controlnet = {}
with safe_open(controlnet_path, framework="pt", device="cpu") as f:
for key in f.keys():
state_dict_controlnet[key] = f.get_tensor(key)
else:
state_dict_controlnet = torch.load(controlnet_path, map_location="cpu")
while "state_dict" in state_dict_controlnet:
state_dict_controlnet = state_dict_controlnet["state_dict"]
self.controlnet = create_diffusers_controlnet_model_from_ldm("ControlNet", self.controlnet_original_config, state_dict_controlnet)['controlnet']
self.controlnet.to(self.dtype)
del state_dict_controlnet
else:
self.controlnet = None
self.current_controlnet_checkpoint = None
# load Adapter_XL
if not hasattr(self, 'adapter'):
adapter_checkpoint_path = os.path.join(script_directory, "checkpoints","X-Adapter")
if not os.path.exists(adapter_checkpoint_path):
try:
from huggingface_hub import snapshot_download
snapshot_download(repo_id="Lingmin-Ran/X-Adapter", local_dir=adapter_checkpoint_path, local_dir_use_symlinks=False)
except:
raise FileNotFoundError(f"No checkpoint directory found at {adapter_checkpoint_path}")
adapter_ckpt = torch.load(os.path.join(adapter_checkpoint_path, "X_Adapter_v1.bin"))
adapter = Adapter_XL()
adapter.load_state_dict(adapter_ckpt)
adapter.to(self.dtype)
# load SDXL
if not hasattr(self, 'unet_sdxl') or self.current_sdxl_checkpoint != sdxl_checkpoint:
self.pipeline = None
comfy.model_management.soft_empty_cache()
print("Loading SDXL checkpoint: ", sdxl_checkpoint)
self.current_sdxl_checkpoint = sdxl_checkpoint
if model_path_sdxl.endswith(".safetensors"):
state_dict_sdxl = {}
with safe_open(model_path_sdxl, framework="pt", device="cpu") as f:
for key in f.keys():
state_dict_sdxl[key] = f.get_tensor(key)
elif model_path_sdxl.endswith(".ckpt"):
state_dict_sdxl = torch.load(model_path_sdxl, map_location="cpu")
while "state_dict" in state_dict_sdxl:
state_dict_sdxl = state_dict_sdxl["state_dict"]
# 1. vae
converted_vae_config = create_vae_diffusers_config(self.sdxl_original_config, image_size=1024)
converted_vae = convert_ldm_vae_checkpoint(state_dict_sdxl, converted_vae_config)
self.vae_sdxl = AutoencoderKL(**converted_vae_config)
self.vae_sdxl.load_state_dict(converted_vae, strict=False)
self.vae_sdxl.to(self.dtype)
# 2. unet
converted_unet_config = create_unet_diffusers_config(self.sdxl_original_config, image_size=1024)
converted_unet = convert_ldm_unet_checkpoint(state_dict_sdxl, converted_unet_config)
self.unet_sdxl = UNet2DConditionModel_v2(**converted_unet_config)
self.unet_sdxl.load_state_dict(converted_unet, strict=False)
self.unet_sdxl.to(self.dtype)
# 3. text encoders and tokenizers
converted_sdxl_stuff = create_text_encoders_and_tokenizers_from_ldm(self.sdxl_original_config, state_dict_sdxl)
self.tokenizer_one = converted_sdxl_stuff['tokenizer']
self.sdxl_text_encoder = converted_sdxl_stuff['text_encoder']
self.tokenizer_two = converted_sdxl_stuff['tokenizer_2']
self.sdxl_text_encoder2 = converted_sdxl_stuff['text_encoder_2']
self.sdxl_text_encoder.to(self.dtype)
self.sdxl_text_encoder2.to(self.dtype)
# 4. scheduler
self.scheduler_sdxl = create_scheduler_from_ldm("DPMSolverMultistepScheduler", self.sdxl_original_config, state_dict_sdxl, scheduler_type="ddim",)['scheduler']
del state_dict_sdxl, converted_unet, converted_sdxl_stuff, converted_vae
#xformers
if use_xformers:
self.unet_sd1_5.enable_xformers_memory_efficient_attention()
self.unet_sdxl.enable_xformers_memory_efficient_attention()
self.controlnet.enable_xformers_memory_efficient_attention()
else:
self.unet_sd1_5.disable_xformers_memory_efficient_attention()
self.unet_sdxl.disable_xformers_memory_efficient_attention()
self.controlnet.disable_xformers_memory_efficient_attention()
self.pipeline = StableDiffusionXLAdapterControlnetPipeline(
vae=self.vae_sdxl,
text_encoder=self.sdxl_text_encoder,
text_encoder_2=self.sdxl_text_encoder2,
tokenizer=self.tokenizer_one,
tokenizer_2=self.tokenizer_two,
unet=self.unet_sdxl,
scheduler=self.scheduler_sdxl,
vae_sd1_5=self.vae_sd1_5,
text_encoder_sd1_5=self.text_encoder_sd1_5,
tokenizer_sd1_5=self.tokenizer_sd1_5,
unet_sd1_5=self.unet_sd1_5,
scheduler_sd1_5=self.scheduler_sd1_5,
adapter=adapter,
controlnet=self.controlnet)
self.pipeline.enable_model_cpu_offload()
self.pipeline.scheduler_sd1_5.config.timestep_spacing = "leading"
#self.pipeline.scheduler.config.timestep_spacing = "trailing"
self.pipeline.unet.to(device=self.device, dtype=self.dtype)
if controlnet_image is not None:
control_image = controlnet_image.permute(0, 3, 1, 2)
else:
control_image = None
width_sdxl = resolution_multiplier * width_sd1_5
height_sdxl = resolution_multiplier * height_sd1_5
#run inference
gen = Generator(self.device)
gen.manual_seed(seed)
img = \
self.pipeline(prompt=prompt_sdxl, negative_prompt=negative_prompt, prompt_sd1_5=prompt_sd1_5,
width=width_sdxl, height=height_sdxl, height_sd1_5=height_sd1_5, width_sd1_5=width_sd1_5,
image=control_image,
num_inference_steps=steps, guidance_scale=cfg,
num_images_per_prompt=batch_size, generator=gen,
controlnet_conditioning_scale=controlnet_condition_scale,
adapter_condition_scale=adapter_condition_scale,
adapter_guidance_start=adapter_guidance_start, guess_mode=guess_mode, control_guidance_start=control_guidance_start,
control_guidance_end=control_guidance_end, source_img=latent_source_image).images
image_tensor = (img - img.min()) / (img.max() - img.min())
if image_tensor.dim() == 3:
image_tensor = image_tensor.unsqueeze(0)
image_tensor = image_tensor.permute(0, 2, 3, 1)
return (image_tensor,)
NODE_CLASS_MAPPINGS = {
"Diffusers_X_Adapter": Diffusers_X_Adapter,
}
NODE_DISPLAY_NAME_MAPPINGS = {
"Diffusers_X_Adapter": "Diffusers_X_Adapter",
}