lee-t/ControlNet-Video
0
1from __future__ import annotations2import gradio as gr3import os4import cv25import numpy as np6from PIL import Image7from moviepy.editor import *8from share_btn import community_icon_html, loading_icon_html, share_js9 10import pathlib11import shlex12import subprocess13 14if os.getenv('SYSTEM') == 'spaces':15 with open('patch') as f:16 subprocess.run(shlex.split('patch -p1'), stdin=f, cwd='ControlNet')17 18base_url = 'https://huggingface.co/lllyasviel/ControlNet/resolve/main/annotator/ckpts/'19 20names = [21 'body_pose_model.pth',22 'dpt_hybrid-midas-501f0c75.pt',23 'hand_pose_model.pth',24 'mlsd_large_512_fp32.pth',25 'mlsd_tiny_512_fp32.pth',26 'network-bsds500.pth',27 'upernet_global_small.pth',28]29 30for name in names:31 command = f'wget https://huggingface.co/lllyasviel/ControlNet/resolve/main/annotator/ckpts/{name} -O {name}'32 out_path = pathlib.Path(f'ControlNet/annotator/ckpts/{name}')33 if out_path.exists():34 continue35 subprocess.run(shlex.split(command), cwd='ControlNet/annotator/ckpts/')36 37from model import (DEFAULT_BASE_MODEL_FILENAME, DEFAULT_BASE_MODEL_REPO,38 DEFAULT_BASE_MODEL_URL, Model)39 40model = Model()41 42 43def controlnet(i, prompt, control_task, seed_in, ddim_steps, scale, low_threshold, high_threshold, value_threshold, distance_threshold, bg_threshold):44 img= Image.open(i)45 np_img = np.array(img)46 47 a_prompt = "best quality, extremely detailed"48 n_prompt = "longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality"49 num_samples = 150 image_resolution = 51251 detect_resolution = 51252 eta = 0.053 #low_threshold = 10054 #high_threshold = 20055 #value_threshold = 0.156 #distance_threshold = 0.157 #bg_threshold = 0.458 59 if control_task == 'Canny':60 result = model.process_canny(np_img, prompt, a_prompt, n_prompt, num_samples,61 image_resolution, ddim_steps, scale, seed_in, eta, low_threshold, high_threshold)62 elif control_task == 'Depth':63 result = model.process_depth(np_img, prompt, a_prompt, n_prompt, num_samples,64 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta)65 elif control_task == 'Hed':66 result = model.process_hed(np_img, prompt, a_prompt, n_prompt, num_samples,67 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta)68 elif control_task == 'Hough':69 result = model.process_hough(np_img, prompt, a_prompt, n_prompt, num_samples,70 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta, value_threshold,71 distance_threshold)72 elif control_task == 'Normal':73 result = model.process_normal(np_img, prompt, a_prompt, n_prompt, num_samples,74 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta, bg_threshold)75 elif control_task == 'Pose':76 result = model.process_pose(np_img, prompt, a_prompt, n_prompt, num_samples,77 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta)78 elif control_task == 'Scribble':79 result = model.process_scribble(np_img, prompt, a_prompt, n_prompt, num_samples,80 image_resolution, ddim_steps, scale, seed_in, eta)81 elif control_task == 'Seg':82 result = model.process_seg(np_img, prompt, a_prompt, n_prompt, num_samples,83 image_resolution, detect_resolution, ddim_steps, scale, seed_in, eta)84 85 #print(result[0])86 processor_im = Image.fromarray(result[0])87 processor_im.save("process_" + control_task + "_" + str(i) + ".jpeg")88 im = Image.fromarray(result[1])89 im.save("your_file" + str(i) + ".jpeg")90 return "your_file" + str(i) + ".jpeg", "process_" + control_task + "_" + str(i) + ".jpeg"91 92def change_task_options(task):93 if task == "Canny" :94 return canny_opt.update(visible=True), hough_opt.update(visible=False), normal_opt.update(visible=False)95 elif task == "Hough" :96 return canny_opt.update(visible=False),hough_opt.update(visible=True), normal_opt.update(visible=False)97 elif task == "Normal" :98 return canny_opt.update(visible=False),hough_opt.update(visible=False), normal_opt.update(visible=True)99 else :100 return canny_opt.update(visible=False),hough_opt.update(visible=False), normal_opt.update(visible=False)101 102def get_frames(video_in):103 frames = []104 #resize the video105 clip = VideoFileClip(video_in)106 107 #check fps108 if clip.fps > 30:109 print("vide rate is over 30, resetting to 30")110 clip_resized = clip.resize(height=512)111 clip_resized.write_videofile("video_resized.mp4", fps=30)112 else:113 print("video rate is OK")114 clip_resized = clip.resize(height=512)115 clip_resized.write_videofile("video_resized.mp4", fps=clip.fps)116 117 print("video resized to 512 height")118 119 # Opens the Video file with CV2120 cap= cv2.VideoCapture("video_resized.mp4")121 122 fps = cap.get(cv2.CAP_PROP_FPS)123 print("video fps: " + str(fps))124 i=0125 while(cap.isOpened()):126 ret, frame = cap.read()127 if ret == False:128 break129 cv2.imwrite('kang'+str(i)+'.jpg',frame)130 frames.append('kang'+str(i)+'.jpg')131 i+=1132 133 cap.release()134 cv2.destroyAllWindows()135 print("broke the video into frames")136 137 return frames, fps138 139 140def convert(gif):141 if gif != None:142 clip = VideoFileClip(gif.name)143 clip.write_videofile("my_gif_video.mp4")144 return "my_gif_video.mp4"145 else:146 pass147 148 149def create_video(frames, fps, type):150 print("building video result")151 clip = ImageSequenceClip(frames, fps=fps)152 clip.write_videofile(type + "_result.mp4", fps=fps)153 154 return type + "_result.mp4"155 156 157def infer(prompt,video_in, control_task, seed_in, trim_value, ddim_steps, scale, low_threshold, high_threshold, value_threshold, distance_threshold, bg_threshold, gif_import):158 print(f"""159 ———————————————160 {prompt}161 ———————————————""")162 163 # 1. break video into frames and get FPS164 break_vid = get_frames(video_in)165 frames_list= break_vid[0]166 fps = break_vid[1]167 n_frame = int(trim_value*fps)168 169 if n_frame >= len(frames_list):170 print("video is shorter than the cut value")171 n_frame = len(frames_list)172 173 # 2. prepare frames result arrays174 processor_result_frames = []175 result_frames = []176 print("set stop frames to: " + str(n_frame))177 178 for i in frames_list[0:int(n_frame)]:179 controlnet_img = controlnet(i, prompt,control_task, seed_in, ddim_steps, scale, low_threshold, high_threshold, value_threshold, distance_threshold, bg_threshold)180 #images = controlnet_img[0]181 #rgb_im = images[0].convert("RGB")182 183 # exporting the image184 #rgb_im.save(f"result_img-{i}.jpg")185 processor_result_frames.append(controlnet_img[1])186 result_frames.append(controlnet_img[0])187 print("frame " + i + "/" + str(n_frame) + ": done;")188 189 processor_vid = create_video(processor_result_frames, fps, "processor")190 final_vid = create_video(result_frames, fps, "final")191 192 files = [processor_vid, final_vid]193 if gif_import != None:194 final_gif = VideoFileClip(final_vid)195 final_gif.write_gif("final_result.gif")196 final_gif = "final_result.gif"197 198 files.append(final_gif)199 print("finished !")200 201 return final_vid, gr.Accordion.update(visible=True), gr.Video.update(value=processor_vid, visible=True), gr.File.update(value=files, visible=True), gr.Group.update(visible=True)202 203 204def clean():205 return gr.Accordion.update(visible=False),gr.Video.update(value=None, visible=False), gr.Video.update(value=None), gr.File.update(value=None, visible=False), gr.Group.update(visible=False)206 207title = """208 <div style="text-align: center; max-width: 700px; margin: 0 auto;">209 <div210 style="211 display: inline-flex;212 align-items: center;213 gap: 0.8rem;214 font-size: 1.75rem;215 "216 >217 <h1 style="font-weight: 900; margin-bottom: 7px; margin-top: 5px;">218 ControlNet Video219 </h1>220 </div>221 <p style="margin-bottom: 10px; font-size: 94%">222 Apply ControlNet to a video 223 </p>224 </div>225"""226 227article = """228 229 <div class="footer">230 <p>231 Follow <a href="https://twitter.com/fffiloni" target="_blank">Sylvain Filoni</a> for future updates 🤗232 </p>233 </div>234 <div id="may-like-container" style="display: flex;justify-content: center;flex-direction: column;align-items: center;margin-bottom: 30px;">235 <p>You may also like: </p>236 <div id="may-like-content" style="display:flex;flex-wrap: wrap;align-items:center;height:20px;">237 238 <svg height="20" width="148" style="margin-left:4px;margin-bottom: 6px;"> 239 <a href="https://huggingface.co/spaces/fffiloni/Pix2Pix-Video" target="_blank">240 <image href="https://img.shields.io/badge/🤗 Spaces-Pix2Pix_Video-blue" src="https://img.shields.io/badge/🤗 Spaces-Pix2Pix_Video-blue.png" height="20"/>241 </a>242 </svg>243 244 </div>245 246 </div>247 248"""249 250with gr.Blocks(css='style.css') as demo:251 with gr.Column(elem_id="col-container"):252 gr.HTML(title)253 gr.HTML("""254 <a style="display:inline-block" href="https://huggingface.co/spaces/fffiloni/ControlNet-Video?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a> 255 """, elem_id="duplicate-container")256 with gr.Row():257 with gr.Column():258 video_inp = gr.Video(label="Video source", source="upload", type="filepath", elem_id="input-vid")259 video_out = gr.Video(label="ControlNet video result", elem_id="video-output")260 261 with gr.Group(elem_id="share-btn-container", visible=False) as share_group:262 community_icon = gr.HTML(community_icon_html)263 loading_icon = gr.HTML(loading_icon_html)264 share_button = gr.Button("Share to community", elem_id="share-btn")265 266 with gr.Accordion("Detailed results", visible=False) as detailed_result:267 prep_video_out = gr.Video(label="Preprocessor video result", visible=False, elem_id="prep-video-output")268 files = gr.File(label="Files can be downloaded ;)", visible=False)269 270 with gr.Column():271 #status = gr.Textbox()272 273 prompt = gr.Textbox(label="Prompt", placeholder="enter prompt", show_label=True, elem_id="prompt-in")274 275 with gr.Row():276 control_task = gr.Dropdown(label="Control Task", choices=["Canny", "Depth", "Hed", "Hough", "Normal", "Pose", "Scribble", "Seg"], value="Pose", multiselect=False, elem_id="controltask-in")277 seed_inp = gr.Slider(label="Seed", minimum=0, maximum=2147483647, step=1, value=123456, elem_id="seed-in")278 279 with gr.Row():280 trim_in = gr.Slider(label="Cut video at (s)", minimun=1, maximum=5, step=1, value=1)281 282 with gr.Accordion("Advanced Options", open=False):283 with gr.Tab("Diffusion Settings"):284 with gr.Row(visible=False) as canny_opt:285 low_threshold = gr.Slider(label='Canny low threshold', minimum=1, maximum=255, value=100, step=1)286 high_threshold = gr.Slider(label='Canny high threshold', minimum=1, maximum=255, value=200, step=1)287 288 with gr.Row(visible=False) as hough_opt:289 value_threshold = gr.Slider(label='Hough value threshold (MLSD)', minimum=0.01, maximum=2.0, value=0.1, step=0.01)290 distance_threshold = gr.Slider(label='Hough distance threshold (MLSD)', minimum=0.01, maximum=20.0, value=0.1, step=0.01)291 292 with gr.Row(visible=False) as normal_opt:293 bg_threshold = gr.Slider(label='Normal background threshold', minimum=0.0, maximum=1.0, value=0.4, step=0.01)294 295 ddim_steps = gr.Slider(label='Steps', minimum=1, maximum=100, value=20, step=1)296 scale = gr.Slider(label='Guidance Scale', minimum=0.1, maximum=30.0, value=9.0, step=0.1)297 298 with gr.Tab("GIF import"):299 gif_import = gr.File(label="import a GIF instead", file_types=['.gif'])300 gif_import.change(convert, gif_import, video_inp, queue=False)301 302 with gr.Tab("Custom Model"):303 current_base_model = gr.Text(label='Current base model',304 value=DEFAULT_BASE_MODEL_URL)305 with gr.Row():306 with gr.Column():307 base_model_repo = gr.Text(label='Base model repo',308 max_lines=1,309 placeholder=DEFAULT_BASE_MODEL_REPO,310 interactive=True)311 base_model_filename = gr.Text(312 label='Base model file',313 max_lines=1,314 placeholder=DEFAULT_BASE_MODEL_FILENAME,315 interactive=True)316 change_base_model_button = gr.Button('Change base model')317 318 gr.HTML(319 '''<p>You can use other base models by specifying the repository name and filename.<br />320 The base model must be compatible with Stable Diffusion v1.5.</p>''')321 322 change_base_model_button.click(fn=model.set_base_model,323 inputs=[324 base_model_repo,325 base_model_filename,326 ],327 outputs=current_base_model, queue=False)328 329 submit_btn = gr.Button("Generate ControlNet video")330 331 inputs = [prompt,video_inp,control_task, seed_inp, trim_in, ddim_steps, scale, low_threshold, high_threshold, value_threshold, distance_threshold, bg_threshold, gif_import]332 outputs = [video_out, detailed_result, prep_video_out, files, share_group]333 #outputs = [status]334 335 336 gr.HTML(article)337 control_task.change(change_task_options, inputs=[control_task], outputs=[canny_opt, hough_opt, normal_opt], queue=False)338 submit_btn.click(clean, inputs=[], outputs=[detailed_result, prep_video_out, video_out, files, share_group], queue=False)339 submit_btn.click(infer, inputs, outputs)340 share_button.click(None, [], [], _js=share_js)341 342 343 344demo.queue(max_size=12).launch()