File size: 6,772 Bytes
9734b1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from enum import Enum
import gc
import numpy as np
import torch


from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
from diffusers import StableDiffusionInstructPix2PixPipeline, StableDiffusionControlNetPipeline, ControlNetModel, UNet2DConditionModel
from diffusers.schedulers import EulerAncestralDiscreteScheduler, DDIMScheduler

from diffusers.pipelines.text_to_video_synthesis.pipeline_text_to_video_zero import CrossFrameAttnProcessor


import utils
import gradio_utils
import os
on_huggingspace = os.environ.get("SPACE_AUTHOR_NAME") == "PAIR"

from einops import rearrange


class ModelType(Enum):
    ControlNetPose = 5,


class Model:
    def __init__(self, device, dtype, **kwargs):
        self.device = device
        self.dtype = dtype
        self.generator = torch.Generator(device=device)
        self.pipe_dict = {
            ModelType.ControlNetPose: StableDiffusionControlNetPipeline,
        }
        
        self.pipe = None
        self.model_type = None

        self.states = {}
        self.model_name = ""

    def set_model(self, model_type: ModelType, model_id: str, **kwargs):
        if hasattr(self, "pipe") and self.pipe is not None:
            del self.pipe
        torch.cuda.empty_cache()
        gc.collect()
        print('kwargs', kwargs)
        print('device', self.device)
        safety_checker = kwargs.pop('safety_checker', None)
        controlnet = kwargs.pop('controlnet', None)
        self.pipe = self.pipe_dict[model_type].from_pretrained(
            model_id, safety_checker=safety_checker, controlnet=controlnet, torch_dtype=torch.float16).to(self.device)#, torch_dtype=torch.float16).to(self.device)

        self.pipe.unet.set_attn_processor(CrossFrameAttnProcessor(batch_size=2))
        self.pipe.controlnet.set_attn_processor(CrossFrameAttnProcessor(batch_size=2)) 

        self.model_type = model_type
        self.model_name = model_id

    def inference_chunk(self, frame_ids, **kwargs):
        if not hasattr(self, "pipe") or self.pipe is None:
            return

        prompt = np.array(kwargs.pop('prompt'))
        negative_prompt = np.array(kwargs.pop('negative_prompt', ''))
        latents = None
        if 'latents' in kwargs:
            latents = kwargs.pop('latents')[frame_ids]
        if 'image' in kwargs:
            kwargs['image'] = kwargs['image'][frame_ids]
        if 'video_length' in kwargs:
            kwargs['video_length'] = len(frame_ids)
        return self.pipe(prompt=prompt[frame_ids].tolist(),
                         negative_prompt=negative_prompt[frame_ids].tolist(),
                         latents=latents,
                         generator=self.generator,
                         **kwargs)

    def inference(self, **kwargs):
        if not hasattr(self, "pipe") or self.pipe is None:
            return

        seed = kwargs.pop('seed', 0)
        if seed < 0:
            seed = self.generator.seed()
        kwargs.pop('generator', '')

        if 'image' in kwargs:
            f = kwargs['image'].shape[0]
        else:
            f = kwargs['video_length']

        assert 'prompt' in kwargs
        prompt = [kwargs.pop('prompt')] * f
        negative_prompt = [kwargs.pop('negative_prompt', '')] * f

        frames_counter = 0

        # Processing frame_by_frame
        result = []
        for i in range(f):
            frame_ids = [0] + [i]
            self.generator.manual_seed(seed)
            print(f'Processing frame {i + 1} / {f}')
            result.append(self.inference_chunk(frame_ids=frame_ids,
                                                   prompt=prompt,
                                                   negative_prompt=negative_prompt,
                                                   **kwargs).images[1:])
            frames_counter += 1
            if on_huggingspace and frames_counter >= 80:
                break
        result = np.concatenate(result)
        return result

    def process_controlnet_pose(self,
                                video_path,
                                prompt,
                                num_inference_steps=20,
                                controlnet_conditioning_scale=1.0,
                                guidance_scale=9.0,
                                seed=42,
                                eta=0.0,
                                resolution=512,
                                use_cf_attn=True,
                                save_path=None):
        print("Module Pose")
        video_path = gradio_utils.motion_to_video_path(video_path)
        if self.model_type != ModelType.ControlNetPose:
            controlnet = ControlNetModel.from_pretrained(
                "fusing/stable-diffusion-v1-5-controlnet-openpose", torch_dtype=torch.float16)
            self.set_model(ModelType.ControlNetPose,
                           model_id="runwayml/stable-diffusion-v1-5", controlnet=controlnet)
            self.pipe.scheduler = DDIMScheduler.from_config(
                self.pipe.scheduler.config)

        video_path = gradio_utils.motion_to_video_path(
            video_path) if 'Motion' in video_path else video_path

        added_prompt = 'best quality, extremely detailed, HD, ultra-realistic, 8K, HQ, masterpiece, trending on artstation, art, smooth'
        negative_prompts = 'longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer difits, cropped, worst quality, low quality, deformed body, bloated, ugly, unrealistic'

        video, fps = utils.prepare_video(
            video_path, resolution, self.device, self.dtype, False, output_fps=4)
        control = utils.pre_process_pose(
            video, apply_pose_detect=False).to(self.device).to(self.dtype)
        f, _, h, w = video.shape
        self.generator.manual_seed(seed)
        latents = torch.randn((1, 4, h//8, w//8), dtype=self.dtype,
                              device=self.device, generator=self.generator)
        latents = latents.repeat(f, 1, 1, 1)
        result = self.inference(image=control,
                                prompt=prompt + ', ' + added_prompt,
                                height=h,
                                width=w,
                                negative_prompt=negative_prompts,
                                num_inference_steps=num_inference_steps,
                                guidance_scale=guidance_scale,
                                controlnet_conditioning_scale=controlnet_conditioning_scale,
                                eta=eta,
                                latents=latents,
                                seed=seed,
                                output_type='numpy',
                                )
        return utils.create_gif(result, fps, path=save_path)