File size: 16,914 Bytes
e60abae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
import os
from typing import List

import torch
from diffusers import StableDiffusionPipeline
from diffusers.pipelines.controlnet import MultiControlNetModel
from PIL import Image
from safetensors import safe_open
from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection

from .utils import is_torch2_available

if is_torch2_available():
    from .attention_processor import (
        AttnProcessor2_0 as AttnProcessor,
    )
    from .attention_processor import (
        CNAttnProcessor2_0 as CNAttnProcessor,
    )
    from .attention_processor import (
        IPAttnProcessor2_0 as IPAttnProcessor,
    )
else:
    from .attention_processor import AttnProcessor, CNAttnProcessor, IPAttnProcessor
from .resampler import Resampler


class ImageProjModel(torch.nn.Module):
    """Projection Model"""

    def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024, clip_extra_context_tokens=4):
        super().__init__()

        self.cross_attention_dim = cross_attention_dim
        self.clip_extra_context_tokens = clip_extra_context_tokens
        self.proj = torch.nn.Linear(clip_embeddings_dim, self.clip_extra_context_tokens * cross_attention_dim)
        self.norm = torch.nn.LayerNorm(cross_attention_dim)

    def forward(self, image_embeds):
        embeds = image_embeds
        clip_extra_context_tokens = self.proj(embeds).reshape(
            -1, self.clip_extra_context_tokens, self.cross_attention_dim
        )
        clip_extra_context_tokens = self.norm(clip_extra_context_tokens)
        return clip_extra_context_tokens


class MLPProjModel(torch.nn.Module):
    """SD model with image prompt"""
    def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024):
        super().__init__()
        
        self.proj = torch.nn.Sequential(
            torch.nn.Linear(clip_embeddings_dim, clip_embeddings_dim),
            torch.nn.GELU(),
            torch.nn.Linear(clip_embeddings_dim, cross_attention_dim),
            torch.nn.LayerNorm(cross_attention_dim)
        )
        
    def forward(self, image_embeds):
        clip_extra_context_tokens = self.proj(image_embeds)
        return clip_extra_context_tokens


class IPAdapter:
    def __init__(self, sd_pipe, 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

        self.pipe = sd_pipe.to(self.device)
        self.set_ip_adapter()

        # load image encoder
        self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
            self.device, dtype=torch.float16
        )
        self.clip_image_processor = CLIPImageProcessor()
        # image proj model
        self.image_proj_model = self.init_proj()

        self.load_ip_adapter()

    def init_proj(self):
        image_proj_model = ImageProjModel(
            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
            clip_embeddings_dim=self.image_encoder.config.projection_dim,
            clip_extra_context_tokens=self.num_tokens,
        ).to(self.device, dtype=torch.float16)
        return image_proj_model

    def set_ip_adapter(self):
        unet = self.pipe.unet
        attn_procs = {}
        for name in unet.attn_processors.keys():
            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
            if name.startswith("mid_block"):
                hidden_size = unet.config.block_out_channels[-1]
            elif name.startswith("up_blocks"):
                block_id = int(name[len("up_blocks.")])
                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
            elif name.startswith("down_blocks"):
                block_id = int(name[len("down_blocks.")])
                hidden_size = unet.config.block_out_channels[block_id]
            if cross_attention_dim is None:
                attn_procs[name] = AttnProcessor()
            else:
                attn_procs[name] = IPAttnProcessor(
                    hidden_size=hidden_size,
                    cross_attention_dim=cross_attention_dim,
                    scale=1.0,
                    num_tokens=self.num_tokens,
                ).to(self.device, dtype=torch.float16)
        unet.set_attn_processor(attn_procs)
        if hasattr(self.pipe, "controlnet"):
            if isinstance(self.pipe.controlnet, MultiControlNetModel):
                for controlnet in self.pipe.controlnet.nets:
                    controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
            else:
                self.pipe.controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))

    def load_ip_adapter(self):
        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
            state_dict = {"image_proj": {}, "ip_adapter": {}}
            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
                for key in f.keys():
                    if key.startswith("image_proj."):
                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
                    elif key.startswith("ip_adapter."):
                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
        else:
            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
        self.image_proj_model.load_state_dict(state_dict["image_proj"])
        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
        ip_layers.load_state_dict(state_dict["ip_adapter"])

    @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=torch.float16)).image_embeds
        else:
            clip_image_embeds = clip_image_embeds.to(self.device, dtype=torch.float16)
        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(clip_image_embeds))
        return image_prompt_embeds, uncond_image_prompt_embeds

    def set_scale(self, scale):
        for attn_processor in self.pipe.unet.attn_processors.values():
            if isinstance(attn_processor, IPAttnProcessor):
                attn_processor.scale = scale

    def generate(
        self,
        pil_image=None,
        clip_image_embeds=None,
        prompt=None,
        negative_prompt=None,
        scale=1.0,
        num_samples=4,
        seed=None,
        guidance_scale=7.5,
        num_inference_steps=30,
        **kwargs,
    ):
        self.set_scale(scale)

        if pil_image is not None:
            num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
        else:
            num_prompts = clip_image_embeds.size(0)

        if prompt is None:
            prompt = "best quality, high quality"
        if negative_prompt is None:
            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"

        if not isinstance(prompt, List):
            prompt = [prompt] * num_prompts
        if not isinstance(negative_prompt, List):
            negative_prompt = [negative_prompt] * num_prompts

        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(
            pil_image=pil_image, clip_image_embeds=clip_image_embeds
        )
        bs_embed, seq_len, _ = image_prompt_embeds.shape
        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)

        with torch.inference_mode():
            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
                prompt,
                device=self.device,
                num_images_per_prompt=num_samples,
                do_classifier_free_guidance=True,
                negative_prompt=negative_prompt,
            )
            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)

        generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
        images = self.pipe(
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            generator=generator,
            **kwargs,
        ).images

        return images


class IPAdapterXL(IPAdapter):
    """SDXL"""

    def generate(
        self,
        pil_image,
        prompt=None,
        negative_prompt=None,
        scale=1.0,
        num_samples=4,
        seed=None,
        num_inference_steps=30,
        **kwargs,
    ):
        self.set_scale(scale)

        num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)

        if prompt is None:
            prompt = "best quality, high quality"
        if negative_prompt is None:
            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"

        if not isinstance(prompt, List):
            prompt = [prompt] * num_prompts
        if not isinstance(negative_prompt, List):
            negative_prompt = [negative_prompt] * num_prompts

        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
        bs_embed, seq_len, _ = image_prompt_embeds.shape
        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)

        with torch.inference_mode():
            (
                prompt_embeds,
                negative_prompt_embeds,
                pooled_prompt_embeds,
                negative_pooled_prompt_embeds,
            ) = self.pipe.encode_prompt(
                prompt,
                num_images_per_prompt=num_samples,
                do_classifier_free_guidance=True,
                negative_prompt=negative_prompt,
            )
            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)

        generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
        images = self.pipe(
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            num_inference_steps=num_inference_steps,
            generator=generator,
            **kwargs,
        ).images

        return images


class IPAdapterPlus(IPAdapter):
    """IP-Adapter with fine-grained features"""

    def init_proj(self):
        image_proj_model = Resampler(
            dim=self.pipe.unet.config.cross_attention_dim,
            depth=4,
            dim_head=64,
            heads=12,
            num_queries=self.num_tokens,
            embedding_dim=self.image_encoder.config.hidden_size,
            output_dim=self.pipe.unet.config.cross_attention_dim,
            ff_mult=4,
        ).to(self.device, dtype=torch.float16)
        return image_proj_model

    @torch.inference_mode()
    def get_image_embeds(self, pil_image=None, clip_image_embeds=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 = clip_image.to(self.device, dtype=torch.float16)
        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
        uncond_clip_image_embeds = self.image_encoder(
            torch.zeros_like(clip_image), output_hidden_states=True
        ).hidden_states[-2]
        uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
        return image_prompt_embeds, uncond_image_prompt_embeds


class IPAdapterFull(IPAdapterPlus):
    """IP-Adapter with full features"""

    def init_proj(self):
        image_proj_model = MLPProjModel(
            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
            clip_embeddings_dim=self.image_encoder.config.hidden_size,
        ).to(self.device, dtype=torch.float16)
        return image_proj_model


class IPAdapterPlusXL(IPAdapter):
    """SDXL"""

    def init_proj(self):
        image_proj_model = Resampler(
            dim=1280,
            depth=4,
            dim_head=64,
            heads=20,
            num_queries=self.num_tokens,
            embedding_dim=self.image_encoder.config.hidden_size,
            output_dim=self.pipe.unet.config.cross_attention_dim,
            ff_mult=4,
        ).to(self.device, dtype=torch.float16)
        return image_proj_model

    @torch.inference_mode()
    def get_image_embeds(self, pil_image):
        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 = clip_image.to(self.device, dtype=torch.float16)
        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
        uncond_clip_image_embeds = self.image_encoder(
            torch.zeros_like(clip_image), output_hidden_states=True
        ).hidden_states[-2]
        uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
        return image_prompt_embeds, uncond_image_prompt_embeds

    def generate(
        self,
        pil_image,
        prompt=None,
        negative_prompt=None,
        scale=1.0,
        num_samples=4,
        seed=None,
        num_inference_steps=30,
        **kwargs,
    ):
        self.set_scale(scale)

        num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)

        if prompt is None:
            prompt = "best quality, high quality"
        if negative_prompt is None:
            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"

        if not isinstance(prompt, List):
            prompt = [prompt] * num_prompts
        if not isinstance(negative_prompt, List):
            negative_prompt = [negative_prompt] * num_prompts

        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
        bs_embed, seq_len, _ = image_prompt_embeds.shape
        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)

        with torch.inference_mode():
            (
                prompt_embeds,
                negative_prompt_embeds,
                pooled_prompt_embeds,
                negative_pooled_prompt_embeds,
            ) = self.pipe.encode_prompt(
                prompt,
                num_images_per_prompt=num_samples,
                do_classifier_free_guidance=True,
                negative_prompt=negative_prompt,
            )
            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)

        generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
        images = self.pipe(
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            num_inference_steps=num_inference_steps,
            generator=generator,
            **kwargs,
        ).images

        return images