app.py

import gradio as gr
import requests
import time
import json
from contextlib import closing
from websocket import create_connection
from deep_translator import GoogleTranslator
from langdetect import detect
import os
from PIL import Image
import io
import base64
import os
import random
import tempfile
import re
from gradio_client import Client
import moviepy.editor as mp


def animate_img(encoded_string, model):
    url_hg1 = os.getenv("url_hg1")
    url_hg2 = os.getenv("url_hg2")
    
    if model == "Stable Video Diffusion":
        try:   
            r = requests.post("https://stable-video-diffusion.com/api/upload", files={"file": open(encoded_string, 'rb')})
            hash_ = r.json()['hash']
            time.sleep(10)
            c = 0
            while c < 10:
                r2 = requests.get(f"https://stable-video-diffusion.com/result?hash={hash_}")
                source_string = r2.text
                if "Generation has been in progress for" in source_string:
                    time.sleep(15)
                    c += 1
                    continue
                if "Generation has been in progress for" not in source_string:
                    pattern = r'https://storage.stable-video-diffusion.com/([a-f0-9]{32})\.mp4'
                    matches = re.findall(pattern, source_string)
                    sd_video = []
                    for match in matches:
                        sd_video.append(f"https://storage.stable-video-diffusion.com/{match}.mp4")
                    if len(sd_video) != 0:
                        print("s_1")
                        return sd_video[0]
                    else:
                        _ = 1/0
            print("f_1")
        except:
            return None
            #print("2")
            #client1 = Client(url_hg1)
            #result1 = client1.predict(encoded_string, api_name="/resize_image")        
            #client = Client(url_hg1)
            #result = client.predict(result1, 0, True, 1, 15, api_name="/video")
            #res = result[0]['video']
            #print("s_2")
            #return res

    #if model == "AnimateDiff":
    #    client = Client(url_hg2)
    #    result = client.predict(encoded_string, "zoom-out", api_name="/predict")
    #    return result
    
    
def create_video(prompt, model):
    url_sd3 = os.getenv("url_sd3")
    url_sd4 = os.getenv("url_sd4")
    if model == "Stable Video Diffusion":
        try:
            with closing(create_connection(f"{url_sd3}", timeout=120)) as conn:
                conn.send('{"fn_index":3,"session_hash":""}')
                conn.send(f'{{"data":["{prompt}","[deformed | disfigured], poorly drawn, [bad : wrong] anatomy, [extra | missing | floating | disconnected] limb, (mutated hands and fingers), blurry",7.5,"(No style)"],"event_data":null,"fn_index":3,"session_hash":""}}')
                c = 0
                while c < 60:
                    status = json.loads(conn.recv())['msg']
                    if status == 'estimation':
                        c += 1
                        time.sleep(1)
                        continue
                    if status == 'process_starts':
                        break
                photo = json.loads(conn.recv())['output']['data'][0][0]
                base64_string = photo.replace('data:image/jpeg;base64,', '').replace('data:image/png;base64,', '')
            
                image_bytes = base64.b64decode(base64_string)
                #with tempfile.NamedTemporaryFile(delete=False) as temp:
                #    temp.write(image_bytes)
                #    temp_file_path = temp.name
                #    print("cs_1")
        except:
            print("c_2")
            with closing(create_connection(f"{url_sd4}", timeout=120)) as conn:
                conn.send('{"fn_index":0,"session_hash":""}')
                conn.send(f'{{"data":["{prompt}","[deformed | disfigured], poorly drawn, [bad : wrong] anatomy, [extra | missing | floating | disconnected] limb, (mutated hands and fingers), blurry","dreamshaperXL10_alpha2.safetensors [c8afe2ef]",30,"DPM++ 2M Karras",7,1024,1024,-1],"event_data":null,"fn_index":0,"session_hash":""}}')
                conn.recv()
                conn.recv()
                conn.recv()
                conn.recv()
                photo = json.loads(conn.recv())['output']['data'][0]
                base64_string = photo.replace('data:image/jpeg;base64,', '').replace('data:image/png;base64,', '')
                image_bytes = base64.b64decode(base64_string)
                #with tempfile.NamedTemporaryFile(delete=False) as temp:
                #    temp.write(image_bytes)
                #    temp_file_path = temp.name
                #    print("cs_2")
    
        try:
            r = requests.post("https://stable-video-diffusion.com/api/upload", files={"file": image_bytes})
            print(r.text)
            hash_ = r.json()['hash']
            time.sleep(10)
            c = 0
            while c < 10:
                r2 = requests.get(f"https://stable-video-diffusion.com/result?hash={hash_}")
                source_string = r2.text
                if "Generation has been in progress for" in source_string:
                    time.sleep(15)
                    c += 1
                    continue
                if "Generation has been in progress for" not in source_string:
                    pattern = r'https://storage.stable-video-diffusion.com/([a-f0-9]{32})\.mp4'
                    matches = re.findall(pattern, source_string)
                    sd_video = []
                    for match in matches:
                        sd_video.append(f"https://storage.stable-video-diffusion.com/{match}.mp4")
                    print(sd_video[0])
                    if len(sd_video) != 0:
                        return sd_video[0]
                    else:
                        _ = 1/0
        except:
            return None
            #client1 = Client("https://emmadrex-stable-video-diffusion.hf.space")
            #result1 = client1.predict(encoded_string, api_name="/resize_image")        
            #client = Client("https://emmadrex-stable-video-diffusion.hf.space")
            #result = client.predict(result1, 0, True, 1, 15, api_name="/video")
            #return result[0]['video']

    if model == "AnimateDiff":
        data = {"prompt": prompt, "negative_prompt": "EasyNegative"}
        r = requests.post("https://sd.cuilutech.com/sdapi/async/txt2gif", json=data)
        c = 0
        while c < 60:
            r2 = requests.post("https://sd.cuilutech.com/sdapi/get_task_info", json={'task_id': r.json()['data']['task_id']})
            time.sleep(2)
            if r2.json()['data']:
                photo = r2.json()['data']['image_urls'][0]
                break
            c += 1
        image = base64.b64encode(requests.get(photo).content).decode("utf-8")
        
        with tempfile.NamedTemporaryFile(delete=False) as temp:
            temp.write(base64.decodebytes(bytes(image, "utf-8")))
            temp_file_path = temp.name
            
        clip = mp.VideoFileClip(temp_file_path)
        n_im2 = f"{time.time()}"
        temp_file2 = tempfile.NamedTemporaryFile(prefix=f'aaafff{n_im2}', suffix='.mp4', delete=False)
        clip.write_videofile(temp_file2.name)
        return temp_file2.name


def flip_text1(prompt, motion):
    try:
        language = detect(prompt)
        if language == 'ru':
            prompt = GoogleTranslator(source='ru', target='en').translate(prompt)
            print(prompt)
    except:
        prompt = 'video'

    url_video_g = os.getenv("url_video_g")
    url_video_c = os.getenv("url_video_c")

    if motion == "Приближение →←":
        motion = 'zoom in'
    if motion == "Отдаление ←→":
        motion = 'zoom out'
    if motion == "Вверх ↑":
        motion = 'up'
    if motion == "Вниз ↓":
        motion = 'down'
    if motion == "Влево ←":
        motion = 'left'
    if motion == "Вправо →":
        motion = 'right'
    if motion == "По часовой стрелке ⟳":
        motion = 'rotate cw'
    if motion == "Против часовой стрелки ⟲":
        motion = 'rotate ccw'
    
    data = {"prompt": f"{prompt}","image": "null", "denoise": 0.75,"motion": motion}
    r = requests.post(f"{url_video_g}", json=data)
    while True:
        data2 = {"task_id": f"{r.json()['task_id']}"}
        r2 = requests.post(f"{url_video_c}", json=data2)
        time.sleep(3)
        try:
            if r2.json()['status'] == "QUEUED":
                continue
            if r2.json()['status'] == "PROCESSING":
                continue
        except:
            try:
                n_im2 = f"{time.time()}"
                with tempfile.NamedTemporaryFile(prefix=f'aaafff{n_im2}', suffix='.mp4', delete=False) as file:
                    for chunk in r2.iter_content(chunk_size=1024):
                        if chunk:
                            file.write(chunk)
                    return file.name
            except Exception as e:
                print(e)
                break





def flip_text2(encoded_string, prompt, motion):

    url_video_g = os.getenv("url_video_g")
    url_video_c = os.getenv("url_video_c")
    
    try:
        language = detect(prompt)
        if language == 'ru':
            prompt = GoogleTranslator(source='ru', target='en').translate(prompt)
            print(prompt)
    except:
        pass

    if motion == "Приближение →←":
        motion = 'zoom in'
    if motion == "Отдаление ←→":
        motion = 'zoom out'
    if motion == "Вверх ↑":
        motion = 'up'
    if motion == "Вниз ↓":
        motion = 'down'
    if motion == "Влево ←":
        motion = 'left'
    if motion == "Вправо →":
        motion = 'right'
    if motion == "По часовой стрелке ⟳":
        motion = 'rotate cw'
    if motion == "Против часовой стрелки ⟲":
        motion = 'rotate ccw'
 
    with open(encoded_string, "rb") as image_file:
        encoded_string2 = base64.b64encode(image_file.read())
        encoded_string2 = str(encoded_string2).replace("b'", '')

    data = {"prompt": f"{prompt}","image": f"{encoded_string2}","denoise":0.75,"motion": motion}
    r = requests.post(f"{url_video_g}", json=data)
    while True:
        data2 = {"task_id": f"{r.json()['task_id']}"}
        r2 = requests.post(f"{url_video_c}", json=data2)
        time.sleep(3)
        try:
            if r2.json()['status'] == "QUEUED":
                continue
            if r2.json()['status'] == "PROCESSING":
                continue
        except:
            try:
                n_im2 = f"{time.time()}"
                with tempfile.NamedTemporaryFile(prefix=f'aaafff{n_im2}', suffix='.mp4', delete=False) as file:
                    for chunk in r2.iter_content(chunk_size=1024):
                        if chunk:
                            file.write(chunk)
                    return file.name
            except Exception as e:
                print(e)
                break

                



css = """
#generate {
    width: 100%;
    background: #e253dd !important;
    border: none;
    border-radius: 50px;
    outline: none !important;
    color: white;
}
#generate:hover {
    background: #de6bda !important;
    outline: none !important;
    color: #fff;
    }
footer {visibility: hidden !important;}
"""

with gr.Blocks(css=css) as demo:

    with gr.Tab("Сгенерировать видео"):
        with gr.Column():
            prompt = gr.Textbox(placeholder="Введите описание видео...", show_label=True, label='Описание:', lines=3)
            # motion1 = gr.Dropdown(value="Приближение →←", interactive=True, show_label=True, label="Движение камеры:", choices=["Приближение →←", "Отдаление ←→", "Вверх ↑", "Вниз ↓", "Влево ←", "Вправо →", "По часовой стрелке ⟳", "Против часовой стрелки ⟲"])
            model = gr.Radio(interactive=True, value="Stable Video Diffusion", show_label=True, 
                             label="Модель нейросети:", choices=['Stable Video Diffusion'])
        with gr.Column():
            text_button = gr.Button("Сгенерировать видео", variant='primary', elem_id="generate")
        with gr.Column():
            video_output = gr.Video(show_label=True, label='Результат:', type="file")
            text_button.click(create_video, inputs=[prompt, model], outputs=video_output)
            
    with gr.Tab("Анимировать изображение"):
        with gr.Column():
            prompt2 = gr.Image(show_label=True, interactive=True, type='filepath', label='Исходное изображение:')
            # prompt12 = gr.Textbox(placeholder="Введите описание видео...", show_label=True, label='Описание видео (опционально):', lines=3)
            # motion2 = gr.Dropdown(value="Приближение →←", interactive=True, show_label=True, label="Движение камеры:", choices=["Приближение →←", "Отдаление ←→", "Вверх ↑", "Вниз ↓", "Влево ←", "Вправо →", "По часовой стрелке ⟳", "Против часовой стрелки ⟲"])
            model2 = gr.Radio(interactive=True, value="Stable Video Diffusion", show_label=True, 
                             label="Модель нейросети:", choices=['Stable Video Diffusion'])
        with gr.Column():
            text_button2 = gr.Button("Анимировать изображение", variant='primary', elem_id="generate")
        with gr.Column():
            video_output2 = gr.Video(show_label=True, label='Результат:', type="file")
            text_button2.click(animate_img, inputs=[prompt2, model2], outputs=video_output2)
    
demo.queue(concurrency_count=12)
demo.launch()