Spaces:
Runtime error
Runtime error
| import os | |
| import time | |
| import gradio as gr | |
| from huggingface_hub import whoami | |
| import threading | |
| from myconfigs import * | |
| from oss_utils import ossService | |
| from gen_client import * | |
| myHumanGen = HumanGenService() | |
| oss_service = ossService() | |
| ENABLE_OSS_RESOURCES = False | |
| profile_name = '' | |
| # result_video_oss_url = {} # user_id, (list[signed_url), list[oss_path])] | |
| def tab_func_template(): | |
| prompt = "" | |
| return prompt, 'template_mode' | |
| def tab_func_prompt(): | |
| ref_video_path = None | |
| return ref_video_path, 'prompt_mode' | |
| def video_2_prompt_func(ref_video_path): | |
| print("[dataset_func] ref_video_path: %s" % ref_video_path) | |
| if not os.path.exists(ref_video_path): | |
| raise gr.Error(f"The input video {ref_video_path} is not existed!") | |
| video_file_name = os.path.basename(ref_video_path) | |
| print("[dataset_func] video_file_name: %s" % video_file_name) | |
| input_prompt = myHumanGen.template_video_2_prompt(video_file_name) | |
| print("[dataset_func] input_prompt: %s" % input_prompt) | |
| return ref_video_path,input_prompt | |
| def get_user_result_video_list(uuid, date_string, num): | |
| directory='video_generation/Service/'+date_string+'/'+uuid+'/' | |
| for obj in oss2.ObjectIterator(oss_service.bucket, prefix=directory, delimiter='/'): | |
| print(f"folder is existed{directory}") | |
| break | |
| else: | |
| print(f"folder is not existed: {directory}") | |
| return [],[] | |
| no_check_video_list = [] | |
| no_check_timer_list = [] | |
| for obj in oss2.ObjectIterator(oss_service.bucket, prefix=directory, delimiter = '/'): | |
| if obj.is_prefix(): # folder | |
| file_full_path = obj.key+'result.mp4' | |
| exist = oss_service.bucket.object_exists(file_full_path) | |
| if not exist: | |
| print(f'{file_full_path} is not existed') | |
| tmp_directory=obj.key | |
| print(f'tmp_directory = {tmp_directory}') | |
| for obj_xxx in oss2.ObjectIterator(oss_service.bucket, prefix=tmp_directory, delimiter = '/'): | |
| print(f'obj_xxx.key = {obj_xxx.key}') | |
| if obj_xxx.is_prefix(): # folder | |
| pass | |
| else: | |
| import re | |
| pattern = r"dreamoving-.*-result\.mp4" | |
| # Extract the MP4 file name | |
| file_name_xxx = os.path.basename(obj_xxx.key) | |
| match = re.search(pattern, obj_xxx.key) | |
| if match and len(match.group()) == len(file_name_xxx): | |
| file_full_path = obj_xxx.key | |
| print(f'successfully match file_full_path: {file_full_path}') | |
| exist = True | |
| break | |
| else: | |
| pass | |
| if exist: | |
| object_meta = oss_service.bucket.head_object(file_full_path) | |
| bytes_num = object_meta.headers.get('Content-Length') # bytes | |
| # print(f"Object Size: {bytes_num} bytes") | |
| mb_num = float(bytes_num) / (1000 ** 2) # MB | |
| # print(f"Object Size: {mb_num} MB") | |
| if mb_num > 0.1: # > 100KB | |
| last_modified = object_meta.headers.get('Last-Modified') | |
| # print(f"Last Modified: {last_modified}") | |
| from email.utils import parsedate_to_datetime | |
| # HTTP-date to datetime | |
| last_modified_datetime = parsedate_to_datetime(last_modified) | |
| # datetime to Unix Time -from 1970-01-01 UTC seconds, nearest is bigger | |
| last_modified_timestamp = int(last_modified_datetime.timestamp()) | |
| no_check_video_list.append(file_full_path) | |
| no_check_timer_list.append(last_modified_timestamp) | |
| else: | |
| print(f'file size: {file_full_path}') | |
| else: # file | |
| print(f'not a file: {obj.key}') | |
| # last_modified = obj.last_modified # last modify time | |
| valid_video_list = [] | |
| valid_image_list = [] | |
| if len(no_check_video_list) > 0: | |
| if len(no_check_video_list) > 1: | |
| # sort by time | |
| zipped_lists = zip(no_check_timer_list, no_check_video_list) | |
| # big to small, nearest is bigger | |
| sorted_pairs = sorted(zipped_lists, key=lambda x: x[0], reverse=True) | |
| list1_sorted, list2_sorted = zip(*sorted_pairs) | |
| no_check_timer_list = list(list1_sorted) | |
| no_check_video_list = list(list2_sorted) | |
| for file_full_path in no_check_video_list: | |
| oss_video_path = oss_service.Prefix + "/" + file_full_path | |
| print(f'Generated video: {oss_video_path}') | |
| _, video_url = oss_service.sign(oss_video_path, timeout=3600*100) | |
| valid_video_list.append(video_url) | |
| style = "video/snapshot,t_1000,f_jpg,w_544,h_768,m_fast" | |
| params1 = {'x-oss-process': style} | |
| _, snapshot_image = oss_service.sign(oss_video_path, timeout=3600*100, params=params1) | |
| valid_image_list.append(snapshot_image) | |
| if len(valid_video_list) >= num: | |
| break | |
| return valid_video_list, valid_image_list | |
| def refresh_video(uuid, request_id): | |
| if uuid is None or uuid == '': | |
| uuid = get_random_string() | |
| print(f'[refresh_video] generate a uuid {uuid}') | |
| # print(f'profile_name====={profile.name}') | |
| notes, process_status = myHumanGen.get_ranking_location(uuid) | |
| print(f'[refresh_video] uuid: {uuid}') | |
| print(f'[refresh_video] request_id: {request_id}') | |
| new_list = [] | |
| new_image_list = [] | |
| if process_status == 'runing': | |
| print(f'process_status: {process_status}') | |
| # new_list.append(None) | |
| # new_image_list.append(None) | |
| date_string = datetime.datetime.now().strftime('%Y-%m-%d') | |
| valid_video_list, valid_image_list = get_user_result_video_list(uuid, date_string, 3) | |
| new_list = new_list + valid_video_list | |
| new_image_list = new_image_list + valid_image_list | |
| if len(new_list) < 4: | |
| date_string_yesterday = (datetime.datetime.now() - datetime.timedelta(days=1)).strftime('%Y-%m-%d') | |
| valid_video_list1, valid_image_list1 = get_user_result_video_list(uuid, date_string_yesterday, 4-len(new_list)) | |
| new_list = new_list + valid_video_list1 | |
| new_image_list = new_image_list + valid_image_list1 | |
| if len(new_list) < 4: | |
| date_string_bf_yesterday = (datetime.datetime.now() - datetime.timedelta(days=2)).strftime('%Y-%m-%d') | |
| valid_video_list2, valid_image_list2 = get_user_result_video_list(uuid, date_string_bf_yesterday, 4-len(new_list)) | |
| new_list = new_list + valid_video_list2 | |
| new_image_list = new_image_list + valid_image_list2 | |
| if len(new_list) < 4: | |
| for i in range(4-len(new_list)): | |
| new_list.append(None) | |
| new_image_list.append(None) | |
| else: | |
| date_string = datetime.datetime.now().strftime('%Y-%m-%d') | |
| valid_video_list, valid_image_list = get_user_result_video_list(uuid, date_string, 4) | |
| new_list = valid_video_list | |
| new_image_list = valid_image_list | |
| if len(new_list) < 4: | |
| date_string_yesterday = (datetime.datetime.now() - datetime.timedelta(days=1)).strftime('%Y-%m-%d') | |
| valid_video_list1, valid_image_list1 = get_user_result_video_list(uuid, date_string_yesterday, 4-len(new_list)) | |
| new_list = new_list + valid_video_list1 | |
| new_image_list = new_image_list + valid_image_list1 | |
| if len(new_list) < 4: | |
| date_string_bf_yesterday = (datetime.datetime.now() - datetime.timedelta(days=2)).strftime('%Y-%m-%d') | |
| valid_video_list2, valid_image_list2 = get_user_result_video_list(uuid, date_string_bf_yesterday, 4-len(new_list)) | |
| new_list = new_list + valid_video_list2 | |
| new_image_list = new_image_list + valid_image_list2 | |
| if len(new_list) < 4: | |
| for i in range(4-len(new_list)): | |
| new_list.append(None) | |
| new_image_list.append(None) | |
| return uuid, notes, new_list[0], new_list[1], new_list[2], new_list[3]#, new_image_list[0], new_image_list[1], new_image_list[2], new_image_list[3] | |
| # def hello(profile: gr.OAuthProfile | None) -> str: | |
| # if profile is None: | |
| # return "I don't know you." | |
| # # print(gr.OAuthProfile) | |
| # profile_name = profile.name | |
| # print(f'profile name={profile.name}') | |
| # return f"Hello {profile.name}" | |
| with gr.Blocks(title = "Dreamoving", | |
| css='style.css', | |
| theme=gr.themes.Soft( | |
| radius_size=gr.themes.sizes.radius_sm, | |
| text_size=gr.themes.sizes.text_md | |
| ) | |
| ) as demo: | |
| # gr.Markdown( | |
| # "# Gradio OAuth Space\n\nThis Space is a demo for the new **Sign in with Hugging Face** feature." | |
| # ) | |
| # gr.LoginButton() | |
| # gr.LogoutButton() | |
| # m_text = gr.Markdown() | |
| # demo.load(hello, inputs=None, outputs=m_text) | |
| with gr.Row(): | |
| gr.HTML(f""" | |
| <div id=css_img_dreamoving> | |
| <img id=css_img_dreamoving src='{RESOURCES.logo_img0}'> | |
| </div> | |
| """) | |
| if ENABLE_OSS_RESOURCES: | |
| template_videos_to_ref = [] | |
| template_video_list = examples['template_video'] | |
| for i in range(9): | |
| file_name = template_video_list[i] | |
| oss_path = oss_service.Prefix + "/" + oss_service.ObjectName + "/template_video/" + file_name | |
| print(f'template_video: {oss_path}') | |
| _, url = oss_service.sign(oss_path, timeout=3600*100) | |
| template_videos_to_ref.append(url) | |
| else: | |
| # template_videos = get_dirnames(filePath="./data/template_video", tail=".mp4") | |
| template_videos_to_ref = [] | |
| template_video_list = examples['template_video'] | |
| for i in range(9): | |
| # file_name = os.path.basename(template_videos[i]) | |
| # file_path = os.path.dirname(template_videos[i]) | |
| file_name = template_video_list[i] | |
| video_path = "./data/template_video/" + file_name | |
| template_videos_to_ref.append(video_path) | |
| # For the same style generation, after users upload a video, they can click the AI button to automatically generate a prompt. | |
| with gr.Accordion(label="🧭 User Guide: It is recommended to read these instructions before using!", open=False): | |
| gr.Markdown(""" | |
| - ⭐️ 1. Video generation time is about 5 minutes. Due to the high number of concurrent users, the generation task may need to queue. Please click the refresh button and check the prompt message. | |
| - ⭐️ 2. If the input image is a cartoon picture, be sure to select "Cartoon Video Generation." | |
| - ⭐️ 3. The system retains up to 4 videos generated in the last two days, refreshing at midnight. Please download and save them in time. | |
| - ⭐️ 4. System updates generally occur between 7-8 a.m. | |
| - ⭐️ 5. Refreshing the page will result in the loss of tasks and generated videos, so please wait for the tasks to be completed and save the generated videos before refreshing. | |
| """) | |
| input_mode = gr.Text(value="template_mode", label="input_mode", visible=False) | |
| with gr.Row(): | |
| with gr.Column(scale=1): | |
| with gr.Group(elem_id='show_box'): | |
| gr.Markdown("Enter/Select a face image") | |
| with gr.Column(): | |
| with gr.Group(elem_id='show_box1'): | |
| with gr.Row(): | |
| ref_image = gr.Image(sources='upload', type='filepath', show_label=False, label='输入图片',elem_id='show_window_image') | |
| gr.Examples(examples['examples_images'], examples_per_page=9, inputs=[ref_image], label='') | |
| with gr.Row(): | |
| model_id = gr.Checkbox(label="Cartoon Video Generation", elem_id='checkbox_0', show_label=False) | |
| with gr.Column(): | |
| gr.Markdown("Select a mode: Reference-Video/Prompt") | |
| with gr.Tab("Guided Style Generation") as tab0: | |
| prompt_template = gr.Textbox(placeholder="Enter prompt words to control the generation effect, such as the character, the character's clothing, the scene, etc.",label="Prompt", lines=2,interactive=True,show_label=False, text_align='left') | |
| # with gr.Row(): | |
| # # with gr.Group(elem_id='show_box3'): | |
| # # with gr.Group(): | |
| # with gr.Column(scale=1, min_width=1): | |
| # prompt_template = gr.Textbox(placeholder="Enter prompt words to control the generation effect, such as the character, the character's clothing, the scene, etc.", label="Prompt提示词", lines=2,interactive=True,show_label=False, text_align='left') | |
| # with gr.Column(scale=1, min_width=1, elem_id='column_button'): | |
| # # prompt_caption_01 = gr.Button(value="AI Caption", elem_id='button_param1') | |
| # prompt_caption_01 = gr.Button( | |
| # value="AI", | |
| # elem_classes='btn_texture_font_file' | |
| # ) | |
| with gr.Row(): | |
| # FIXME: the width/height setting not work here, | |
| ref_video = gr.Video(sources='upload', show_label=False, label='Input Video', autoplay=True, elem_id='show_window_video', width=224, height=360) | |
| # gr.Examples(examples['template_video'], examples_per_page=9,inputs=[ref_video], label='Template Video') | |
| # dataset_select = gr.Dataset( | |
| # label='Template Video', | |
| # components=[gr.Video(visible=False)], | |
| # samples=examples['template_video'], | |
| # samples_per_page=9, | |
| # type='index', # pass index or value | |
| # # min_width=400, | |
| # # elem_id='dataset', | |
| # # elem_id='template_param', | |
| # ) | |
| gr.Examples( | |
| label='Template Video', | |
| examples=template_videos_to_ref, | |
| inputs=ref_video, | |
| outputs=[ref_video, prompt_template], | |
| fn=video_2_prompt_func, | |
| examples_per_page=9, | |
| cache_examples=True, #run_on_click=True, | |
| ) | |
| # prompt_template = gr.Textbox(label="Prompt", lines=2,interactive=True,show_label=False, text_align='left') | |
| with gr.Tab("Text-to-Video") as tab1: | |
| # prompt = gr.Textbox(label="Prompt", show_label=False, text_align='left') | |
| example_prompts= [] | |
| file = open(ref_video_prompt, 'r') | |
| for line in file.readlines(): | |
| example_prompts.append(line) | |
| file.close() | |
| prompt = gr.Dropdown(label="Prompt List",choices=example_prompts, show_label=False, allow_custom_value=True) | |
| with gr.Row(): | |
| # Generate Button | |
| run_button = gr.Button(value="Generation", elem_id='button_param') | |
| # btn = gr.Button("Result Video").style(full_width=False) | |
| with gr.Column(scale=1): | |
| # gr.Markdown("Result Video",elem_id='font_style') | |
| with gr.Group(elem_id='show_box2'): | |
| with gr.Row(elem_id='round_box'): | |
| with gr.Column(scale=1, min_width=1): | |
| gr.Markdown("Result Video", elem_id='font_style') | |
| with gr.Column(scale=3, min_width=1): | |
| user_notes = gr.Textbox(show_label=False, text_align='left', elem_id='text_style11') | |
| with gr.Column(scale=1, min_width=1): | |
| refresh_button = gr.Button(value="Refresh", elem_id='button_param1') | |
| with gr.Row(): | |
| output_video0 = gr.Video(format="mp4", show_label=False, label="Result Video", autoplay=True, elem_id='show_window_result') | |
| output_video1 = gr.Video(format="mp4", show_label=False, label="Result Video", autoplay=True,elem_id='show_window_result') | |
| with gr.Row(): | |
| output_video2 = gr.Video(format="mp4", show_label=False, label="Result Video", autoplay=True,elem_id='show_window_result') | |
| output_video3 = gr.Video(format="mp4", show_label=False, label="Result Video", autoplay=True,elem_id='show_window_result') | |
| uuid = gr.Text(label="hf_uuid", visible=False) | |
| request_id = gr.Text(label="hf_request_id", visible=False) | |
| # Sample Video | |
| num_video = 8 | |
| num_videos_per_row = 4 | |
| mp4_lists = [] | |
| if ENABLE_OSS_RESOURCES: | |
| mp4_url_list = get_dirnames(filePath="./data/sample_video", tail=".mp4") | |
| for i in range(min(num_video, len(mp4_url_list))): | |
| file_name = os.path.basename(mp4_url_list[i]) | |
| oss_path = oss_service.Prefix + "/" + oss_service.ObjectName + "/template_video/" + file_name | |
| print(f'template_video: {oss_path}') | |
| _, video_url = oss_service.sign(oss_path, timeout=3600*100) | |
| mp4_lists.append(video_url) | |
| else: | |
| mp4_lists = get_dirnames(filePath="./data/sample_video", tail=".mp4") | |
| if len(mp4_lists) <= num_video: | |
| num_video = len(mp4_lists) | |
| with gr.Row(): | |
| gr.Markdown("Sample Video",elem_id='font_style') | |
| with gr.Group(elem_id='show_box'): | |
| with gr.Column(): | |
| for i in range(int((num_video+num_videos_per_row-1)/num_videos_per_row)): | |
| with gr.Row(): | |
| for j in range(num_videos_per_row): | |
| if i*num_videos_per_row+j < len(mp4_lists): | |
| gr.Video(value=mp4_lists[i*num_videos_per_row+j], show_label=False, interactive=False, label='result') | |
| else: | |
| gr.Video(interactive=False, label='result') | |
| refresh_button.click( | |
| fn=refresh_video, | |
| queue = False, | |
| inputs=[uuid, request_id], | |
| outputs=[uuid, user_notes, output_video0, output_video1, output_video2, output_video3] | |
| ) | |
| # prompt_caption_01.click( | |
| # fn=myHumanGen.click_button_prompt, | |
| # queue = False, | |
| # inputs=[uuid, request_id, input_mode, ref_image, ref_video, prompt, prompt_template, model_id], | |
| # outputs=[prompt_template] | |
| # ) | |
| # dataset_select.select(fn=dataset_func, outputs=[ref_video,prompt_template]) | |
| # button触发 | |
| tab0.select(fn=tab_func_template, outputs=[prompt, input_mode]) # template mode | |
| tab1.select(fn=tab_func_prompt, outputs=[ref_video, input_mode]) # prompt mode | |
| def async_process(user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt='', prompt_template='',model_id=False): | |
| if user_id is None or user_id == '': | |
| user_id = get_random_string() | |
| print(f'[async_process] generate a uuid {uuid}') | |
| # parm-chheck | |
| check_note_info = myHumanGen.valid_check(user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt, prompt_template,model_id) | |
| if check_note_info == '': | |
| thread = threading.Thread(target=myHumanGen.click_button_func_async, args=(user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt, prompt_template,model_id,)) | |
| thread.start() | |
| # thread.join() | |
| time.sleep(5) | |
| return refresh_video(user_id, request_id) | |
| else: | |
| user_id, notes, video_0, video_1, video_2, video_3 = refresh_video(user_id, request_id) | |
| return user_id, check_note_info, video_0, video_1, video_2, video_3 | |
| run_button.click(fn=async_process, inputs=[uuid, request_id, input_mode, ref_image, ref_video, prompt, prompt_template, model_id], outputs=[uuid, user_notes, output_video0, output_video1, output_video2, output_video3]) | |
| with gr.Row(): | |
| # DingTalk | |
| gr.HTML(f""" | |
| <div id=css_img_QRCode> | |
| <img id=css_img_QRCode src='{RESOURCES.logo_dingding}'> | |
| </div> | |
| <div id=css_img_QRCode_text> | |
| DingTalk Group of Dreamoving | |
| </div> | |
| """) | |
| gr.HTML(f""" | |
| <div id=css_img_QRCode> | |
| <img id=css_img_QRCode src='{RESOURCES.logo_wechat}'> | |
| </div> | |
| <div id=css_img_QRCode_text> | |
| WeChat Group of Dreamoving | |
| </div> | |
| """) | |
| # version | |
| gr.HTML(f""" | |
| </br> | |
| <div> | |
| <center> Version: {VERSION} </center> | |
| </div> | |
| """) | |
| # concurrency_count, concurrency_limit, max_threads | |
| demo.queue(api_open=False, max_size=1000).launch( | |
| server_name="0.0.0.0", # if os.getenv('GRADIO_LISTEN', '') != '' else "127.0.0.1", | |
| share=True, | |
| server_port=7860, | |
| root_path=f"/{os.getenv('GRADIO_PROXY_PATH')}" if os.getenv('GRADIO_PROXY_PATH') else "" | |
| ) | |