-
Notifications
You must be signed in to change notification settings - Fork 17
/
Copy pathipadapter_flux_advanced.py
186 lines (162 loc) · 8.42 KB
/
ipadapter_flux_advanced.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
import torch
import os
import logging
import folder_paths
from transformers import AutoProcessor, SiglipVisionModel
from PIL import Image
import numpy as np
from .attention_processor_advanced import IPAFluxAttnProcessor2_0Advanced
from .utils import is_model_patched, FluxUpdateModules
MODELS_DIR = os.path.join(folder_paths.models_dir, "ipadapter-flux")
if "ipadapter-flux" not in folder_paths.folder_names_and_paths:
current_paths = [MODELS_DIR]
else:
current_paths, _ = folder_paths.folder_names_and_paths["ipadapter-flux"]
folder_paths.folder_names_and_paths["ipadapter-flux"] = (current_paths, folder_paths.supported_pt_extensions)
class MLPProjModelAdvanced(torch.nn.Module):
def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, num_tokens=4):
super().__init__()
self.cross_attention_dim = cross_attention_dim
self.num_tokens = num_tokens
self.proj = torch.nn.Sequential(
torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2),
torch.nn.GELU(),
torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens),
)
self.norm = torch.nn.LayerNorm(cross_attention_dim)
def forward(self, id_embeds):
x = self.proj(id_embeds)
x = x.reshape(-1, self.num_tokens, self.cross_attention_dim)
x = self.norm(x)
return x
class InstantXFluxIPAdapterModelAdvanced:
def __init__(self, image_encoder_path, ip_ckpt, device, num_tokens=4):
self.device = device
self.image_encoder_path = image_encoder_path
self.ip_ckpt = ip_ckpt
self.num_tokens = num_tokens
# load image encoder
self.image_encoder = SiglipVisionModel.from_pretrained(self.image_encoder_path).to(self.device, dtype=torch.float16)
self.clip_image_processor = AutoProcessor.from_pretrained(self.image_encoder_path)
# state_dict
self.state_dict = torch.load(os.path.join(MODELS_DIR,self.ip_ckpt), map_location="cpu")
self.joint_attention_dim = 4096
self.hidden_size = 3072
def init_proj(self):
self.image_proj_model = MLPProjModelAdvanced(
cross_attention_dim=self.joint_attention_dim, # 4096
id_embeddings_dim=1152,
num_tokens=self.num_tokens,
).to(self.device, dtype=torch.float16)
def set_ip_adapter_advanced(self, flux_model, weight_params, timestep_percent_range=(0.0, 1.0)):
weight_start, weight_end, steps = weight_params
s = flux_model.model_sampling
percent_to_timestep_function = lambda a: s.percent_to_sigma(a)
timestep_range = (percent_to_timestep_function(timestep_percent_range[0]), percent_to_timestep_function(timestep_percent_range[1]))
ip_attn_procs = {}
dsb_count = len(flux_model.diffusion_model.double_blocks)
for i in range(dsb_count):
name = f"double_blocks.{i}"
ip_attn_procs[name] = IPAFluxAttnProcessor2_0Advanced(
hidden_size=self.hidden_size,
cross_attention_dim=self.joint_attention_dim,
num_tokens=self.num_tokens,
scale_start=weight_start,
scale_end=weight_end,
total_steps=steps,
timestep_range=timestep_range
).to(self.device, dtype=torch.float16)
ssb_count = len(flux_model.diffusion_model.single_blocks)
for i in range(ssb_count):
name = f"single_blocks.{i}"
ip_attn_procs[name] = IPAFluxAttnProcessor2_0Advanced(
hidden_size=self.hidden_size,
cross_attention_dim=self.joint_attention_dim,
num_tokens=self.num_tokens,
scale_start=weight_start,
scale_end=weight_end,
total_steps=steps,
timestep_range=timestep_range
).to(self.device, dtype=torch.float16)
return ip_attn_procs
def load_ip_adapter_advanced(self, flux_model, weight, timestep_percent_range=(0.0, 1.0)):
self.image_proj_model.load_state_dict(self.state_dict["image_proj"], strict=True)
ip_attn_procs = self.set_ip_adapter_advanced(flux_model, weight, timestep_percent_range)
ip_layers = torch.nn.ModuleList(ip_attn_procs.values())
ip_layers.load_state_dict(self.state_dict["ip_adapter"], strict=True)
return ip_attn_procs
@torch.inference_mode()
def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
if pil_image is not None:
if isinstance(pil_image, Image.Image):
pil_image = [pil_image]
clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
clip_image_embeds = self.image_encoder(clip_image.to(self.device, dtype=self.image_encoder.dtype)).pooler_output
clip_image_embeds = clip_image_embeds.to(dtype=torch.float16)
else:
clip_image_embeds = clip_image_embeds.to(self.device, dtype=torch.float16)
image_prompt_embeds = self.image_proj_model(clip_image_embeds)
return image_prompt_embeds
class IPAdapterFluxLoaderAdvanced:
@classmethod
def INPUT_TYPES(s):
return {"required": {
"ipadapter": (folder_paths.get_filename_list("ipadapter-flux"),),
"clip_vision": (["google/siglip-so400m-patch14-384"],),
"provider": (["cuda", "cpu", "mps"],),
}
}
RETURN_TYPES = ("IP_ADAPTER_FLUX_INSTANTX",)
RETURN_NAMES = ("ipadapterFlux",)
FUNCTION = "load_model_advanced"
CATEGORY = "InstantXNodes"
def load_model_advanced(self, ipadapter, clip_vision, provider):
logging.info("Loading InstantX IPAdapter Flux model.")
model = InstantXFluxIPAdapterModelAdvanced(image_encoder_path=clip_vision, ip_ckpt=ipadapter, device=provider, num_tokens=128)
return (model,)
class ApplyIPAdapterFluxAdvanced:
@classmethod
def INPUT_TYPES(s):
return {
"required": {
"model": ("MODEL", ),
"ipadapter_flux": ("IP_ADAPTER_FLUX_INSTANTX", ),
"image": ("IMAGE", ),
"weight_start": ("FLOAT", {"default": 0.0, "min": -1.0, "max": 5.0, "step": 0.05 }),
"weight_end": ("FLOAT", {"default": 1.0, "min": -1.0, "max": 5.0, "step": 0.05 }),
"steps": ("INT", {"default": 10, "min": 1, "max": 100, "step": 1}),
"start_percent": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.001}),
"end_percent": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.001})
},
}
RETURN_TYPES = ("MODEL",)
FUNCTION = "apply_ipadapter_flux_advanced"
CATEGORY = "InstantXNodes"
def apply_ipadapter_flux_advanced(self, model, ipadapter_flux, image, weight_start, weight_end, steps, start_percent, end_percent):
# Clean up old processors if they exist
if hasattr(model.model, '_ip_attn_procs'):
for proc in model.model._ip_attn_procs.values():
proc.clear_memory() # Add a new method for cleanup
del model.model._ip_attn_procs
pil_image = image.numpy()[0] * 255.0
pil_image = Image.fromarray(pil_image.astype(np.uint8))
ipadapter_flux.init_proj()
IPAFluxAttnProcessor2_0Advanced.reset_all_instances()
ip_attn_procs = ipadapter_flux.load_ip_adapter_advanced(model.model, (weight_start, weight_end, steps), (start_percent, end_percent))
image_prompt_embeds = ipadapter_flux.get_image_embeds(
pil_image=pil_image, clip_image_embeds=None
)
is_patched = is_model_patched(model.model)
bi = model.clone()
FluxUpdateModules(bi, ip_attn_procs, image_prompt_embeds, is_patched)
# Store reference to processors for cleanup
bi.model._ip_attn_procs = ip_attn_procs
return (bi,)
NODE_CLASS_MAPPINGS = {
"IPAdapterFluxLoaderAdvanced": IPAdapterFluxLoaderAdvanced,
"ApplyIPAdapterFluxAdvanced": ApplyIPAdapterFluxAdvanced,
}
NODE_DISPLAY_NAME_MAPPINGS = {
"IPAdapterFluxLoaderAdvanced": "Load IPAdapter Flux Model (Advanced)",
"ApplyIPAdapterFluxAdvanced": "Apply IPAdapter Flux Model (Advanced)",
}