Spaces:
Running
Running
| import os | |
| import requests | |
| import gradio as gr | |
| import uuid | |
| import datetime | |
| from supabase import create_client, Client | |
| from supabase.lib.client_options import ClientOptions | |
| import dotenv | |
| from google.cloud import storage | |
| import json | |
| from pathlib import Path | |
| import mimetypes | |
| from workflow_handler import WanVideoWorkflow | |
| from video_config import MODEL_FRAME_RATES, calculate_frames | |
| import asyncio | |
| from openai import OpenAI | |
| import base64 | |
| from google.cloud import vision | |
| from google.oauth2 import service_account | |
| dotenv.load_dotenv() | |
| SCRIPT_DIR = Path(__file__).parent | |
| CONFIG_PATH = SCRIPT_DIR / "config.json" | |
| WORKFLOW_PATH = SCRIPT_DIR / "wani2v.json" | |
| loras = [ | |
| { | |
| #I suggest it to be a gif instead of an mp4! | |
| "image": "https://huggingface.co/Remade-AI/Squish/resolve/main/example_gifs/person_squish.gif", | |
| #This is an id you can send to your backend, obviously you can change it | |
| "id": "06ce6840-f976-4963-9644-b6cf7f323f90", | |
| #This is the title that is shown on the front end | |
| "title": "Squish", | |
| "example_prompt": "In the video, a miniature rodent is presented. The rodent is held in a person's hands. The person then presses on the rodent, causing a sq41sh squish effect. The person keeps pressing down on the rodent, further showing the sq41sh squish effect.", | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Rotate/resolve/main/example_videos/chair-rotate.gif", | |
| "id": "4ac08cfa-841e-4aa9-9022-c3fc80fb6ef4", | |
| "title": "Rotate", | |
| "example_prompt": "The video shows an elderly Asian man's head and shoulders with blurred background, performing a r0t4tion 360 degrees rotation.", | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Cakeify/resolve/main/example_gifs/timberland_cakeify.gif", | |
| "id": "b05c1dc7-a71c-4d24-b512-4877a12dea7e", | |
| "title": "Cakeify", | |
| "example_prompt": "The video opens on a woman. A knife, held by a hand, is coming into frame and hovering over the woman. The knife then begins cutting into the woman to c4k3 cakeify it. As the knife slices the woman open, the inside of the woman is revealed to be cake with chocolate layers. The knife cuts through and the contents of the woman are revealed." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Muscle/resolve/main/example_videos/man2_muscle.gif", | |
| "id": "3c6fd399-e558-43fa-8cd3-828300aac6f8", | |
| "title": "Muscle", | |
| "example_prompt": "A man t2k1s takes off clothes revealing a lean muscular body and shows off muscles, looking towards the camera." | |
| }, | |
| { | |
| "image": "https://storage.googleapis.com/remade-v2/huggingface_assets/crush_example.gif", | |
| "id": "d8a2912b-94e4-4227-9c45-356679af34fd", | |
| "title": "Crush", | |
| "example_prompt": "The video begins with a cube saying closed source. A hydraulic press positioned above slowly descends towards the cube. Upon contact, the hydraulic press c5us4 crushes it, deforming and flattening the cube, causing the cube to collapse inward until the cube is no longer recognizable." | |
| }, | |
| { | |
| "image": "https://storage.googleapis.com/remade-v2/huggingface_assets/decay_example.gif", | |
| "id": "6b6f64dc-ac14-44b2-b91c-a510cb7f7f32", | |
| "title": "Decay", | |
| "example_prompt": "The video shows a man. The d3c4y decay time-lapse begins, causing the man to change. The man is initially whole, but soon he appears to be rotting. The man slowly becomes increasingly shriveled and discolored, and eventually, the man decomposes and falls apart. The man is rotting in the center and appears to be covered in mold, completing the d3c4y decay time-lapse." | |
| }, | |
| { | |
| "image": "https://storage.googleapis.com/remade-v2/huggingface_assets/jesus_example.gif", | |
| "id": "615fe106-fec4-44bb-b28b-2864cb322027", | |
| "title": "Jesus", | |
| "example_prompt": "The video begins with a smiling woman with a pink shirt looking at the camera. Then jesus appears behind her as h54g hugs jesus. Jesus embraces the woman, and they both smile in front of a park." | |
| }, | |
| { | |
| "image": "https://storage.googleapis.com/remade-v2/huggingface_assets/inflate_example.gif", | |
| "id": "da2b1c34-7be8-4161-a733-e8b19a98901c", | |
| "title": "Inflate", | |
| "example_prompt": "The large, bald man rides a gray donkey, then infl4t3 inflates it, both the man and the donkey expanding into giant, inflated figures against the desert landscape." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Jungle/resolve/main/example_videos/man1_jungle.gif", | |
| "id": "cf749aeb-5f25-4c6e-b495-3ea8d81004ee", | |
| "title": "Jungle It", | |
| "example_prompt": "The video begins with a portrait of a man. The background is blurry, with shades of grey and green. Next, the 1ung13 jungle transformation occurs. The man is now in a jungle setting, bathed in sunlight. His hair is longer, and his hair is up. He is shirtless, with tribal markings on his chest. He wears jungle-like shorts. The man is swinging from a vine, posing in a dynamic, action-oriented manner. A dark panther-like figure is in the background. The scene evokes a sense of adventure and the wild." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Baby/resolve/main/example_videos/goku_baby.gif", | |
| "id": "5e45b11e-b9ff-404a-9afa-22a3c5596523c", | |
| "title": "Baby It", | |
| "example_prompt": "The video starts with a studio portrait of a woman. Then the image shifts to the 848y baby effect, the woman is in front of a crib, surrounded by toys. Finally, the 848y baby effect is shown again in a different location. The 848y baby version of the woman is in the crib and seems excited and amused." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Assassin/resolve/main/example_videos/dog_assassin.gif", | |
| "id": "88600b53-336a-4d0c-a1a4-8b53e9775f03", | |
| "title": "Assassin It", | |
| "example_prompt": "The video starts with a portrait of a dog. Then, the 3p1c epic transformation starts. The dog is wearing a red coat, and the 3p1c epic transformation is complete. The dog is holding a gun in each hand. The dog has white hair and black gloves." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Warrior/resolve/main/example_videos/dog_warrior.gif", | |
| "id": "4140f3c2-430d-4b47-b40a-997f361d83dc", | |
| "title": "Warrior It", | |
| "example_prompt": "The video starts with a woman. The next scene shows her with a mountain range in the background. her shirt is pulled up to his midriff, and she is wearing a skirt-like bottom. The woman has a belt around her waist and is gesturing with her right hand. She is wearing brown, medieval looking leggings. The effect seen is warr10r warrior it. The woman now appears as a warrior with an axe. She is shirtless, muscular, has tattoos, and is smiling with a determined look on her face. The background is the same mountain range as before. The next scene shows the woman still as a warrior, and he is holding an axe with a golden axe head." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Pirate-Captain/resolve/main/example_videos/cat_example.gif.gif", | |
| "id": "26c4248e-4289-4964-b01b-ace89c7ad407", | |
| "title": "Pirate It", | |
| "example_prompt": "The video begins with a man posing. The image then transitions to the p1r4t3 pirate captain transformation. The man is wearing a black pirate hat with a red band around it, a coat and pants, and a pirate style sash. The scene changes, showing the man on a wooden ship. He has long dreadlock style hair and a sword. The scene changes again to show the man with his sword, in the same location on the boat." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Bride/resolve/main/example_videos/rabbit_bride.gif", | |
| "id": "bd3100fe-65be-416c-994f-bb5acee1404d", | |
| "title": "Bride It", | |
| "example_prompt": "The video begins with a portrait of a bunny rabbit, then the 8r1d3 bride effect occurs. The bunny rabbit is now in a white wedding dress, holding a bouquet, with a sunny, warm beige background. " | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/VIP/resolve/main/example_videos/thanos_vip.gif", | |
| "id": "fa3355bc-2b7c-42f6-b22e-a6b07937a20c", | |
| "title": "VIP it", | |
| "example_prompt": "The video begins with an image of purple Thanos from Marvel. Then the v1p red carpet transformation appears. Purple Thanos is shown wearing a black dress, with gold jewelry around his neck and ears. The image is again of purple Thanos looking straight at the camera against a more lighted gray background. The v1p red carpet transformation continues, purple Thanos is now on the red carpet with photographers taking pictures and other people behind a barricade to the sides. Purple Thanos is wearing the same black dress and jewelry, in focus at the center of the frame." | |
| }, | |
| { | |
| "image": "https://huggingface.co/Remade-AI/Zen/resolve/main/example_videos/man_zen.gif", | |
| "id": "328c6078-515a-4fa0-8b5d-9ea993954f80", | |
| "title": "Zen It", | |
| "example_prompt": "The video starts with a portrait of a purple Thanos from Marvel. The scene then transitions to the Thanos' z3n1fy zen transformation as he's wearing a pink robe with a white shirt underneath, with a zen garden background. Thanos is facing the camera with a neutral expression. The background appears to be blurred and out of focus. The scene then transitions again to show the transformed Thanos, in what appears to be a garden setting. He is now wearing a black kimono with white floral designs and a white belt. Thanos carries a basket in one hand and a colorful fan in the other. He is walking down a pathway lined with hedges and greenery. The z3n1fy zen transformation is complete. Thanos has a neutral expression, looking directly at the camera." | |
| }, | |
| ] | |
| # Initialize Supabase client with async support | |
| supabase: Client = create_client( | |
| os.getenv('SUPABASE_URL'), | |
| os.getenv('SUPABASE_KEY'), | |
| ) | |
| # Initialize OpenAI client | |
| openai_client = OpenAI(api_key=os.getenv('OPENAI_API_KEY')) | |
| def initialize_gcs(): | |
| """Initialize Google Cloud Storage client with credentials from environment""" | |
| try: | |
| # Parse service account JSON from environment variable | |
| service_account_json = os.getenv('SERVICE_ACCOUNT_JSON') | |
| if not service_account_json: | |
| raise ValueError("SERVICE_ACCOUNT_JSON environment variable not found") | |
| credentials_info = json.loads(service_account_json) | |
| # Initialize storage client | |
| storage_client = storage.Client.from_service_account_info(credentials_info) | |
| print("Successfully initialized Google Cloud Storage client") | |
| return storage_client | |
| except Exception as e: | |
| print(f"Error initializing Google Cloud Storage: {e}") | |
| raise | |
| def upload_to_gcs(file_path, content_type=None, folder='user_uploads'): | |
| """ | |
| Uploads a file to Google Cloud Storage | |
| Args: | |
| file_path: Path to the file to upload | |
| content_type: MIME type of the file (optional) | |
| folder: Folder path in bucket (default: 'user_uploads') | |
| Returns: | |
| str: Public URL of the uploaded file | |
| """ | |
| try: | |
| bucket_name = 'remade-v2' | |
| storage_client = initialize_gcs() | |
| bucket = storage_client.bucket(bucket_name) | |
| # Get file extension and generate unique filename | |
| file_extension = Path(file_path).suffix | |
| if not content_type: | |
| content_type = mimetypes.guess_type(file_path)[0] or 'application/octet-stream' | |
| # Validate file type | |
| valid_types = ['image/jpeg', 'image/png', 'image/gif'] | |
| if content_type not in valid_types: | |
| raise ValueError("Invalid file type. Please upload a JPG, PNG or GIF image.") | |
| # Generate unique filename with proper path structure | |
| filename = f"{str(uuid.uuid4())}{file_extension}" | |
| file_path_in_gcs = f"{folder}/{filename}" | |
| # Create blob and set metadata | |
| blob = bucket.blob(file_path_in_gcs) | |
| blob.content_type = content_type | |
| blob.cache_control = 'public, max-age=31536000' | |
| print(f'Uploading file to GCS: {file_path_in_gcs}') | |
| # Upload the file | |
| blob.upload_from_filename( | |
| file_path, | |
| timeout=120 # 2 minute timeout | |
| ) | |
| # Generate public URL with correct path format | |
| image_url = f"https://storage.googleapis.com/{bucket_name}/{file_path_in_gcs}" | |
| print(f"Successfully uploaded to GCS: {image_url}") | |
| return image_url | |
| except Exception as e: | |
| print(f"Error uploading to GCS: {e}") | |
| raise ValueError(f"Failed to upload image to storage: {str(e)}") | |
| def build_lora_prompt(subject, lora_id): | |
| """ | |
| Builds a standardized prompt based on the selected LoRA and subject | |
| """ | |
| # Get LoRA config | |
| lora_config = next((lora for lora in loras if lora["id"] == lora_id), None) | |
| if not lora_config: | |
| raise ValueError(f"Invalid LoRA ID: {lora_id}") | |
| if lora_id == "06ce6840-f976-4963-9644-b6cf7f323f90": # Squish | |
| return ( | |
| f"In the video, a miniature {subject} is presented. " | |
| f"The {subject} is held in a person's hands. " | |
| f"The person then presses on the {subject}, causing a sq41sh squish effect. " | |
| f"The person keeps pressing down on the {subject}, further showing the sq41sh squish effect." | |
| ) | |
| elif lora_id == "4ac08cfa-841e-4aa9-9022-c3fc80fb6ef4": # Rotate | |
| return ( | |
| f"The video shows a {subject} performing a r0t4tion 360 degrees rotation." | |
| ) | |
| elif lora_id == "b05c1dc7-a71c-4d24-b512-4877a12dea7e": # Cakeify | |
| return ( | |
| f"The video opens on a {subject}. A knife, held by a hand, is coming into frame " | |
| f"and hovering over the {subject}. The knife then begins cutting into the {subject} " | |
| f"to c4k3 cakeify it. As the knife slices the {subject} open, the inside of the " | |
| f"{subject} is revealed to be cake with chocolate layers. The knife cuts through " | |
| f"and the contents of the {subject} are revealed." | |
| ) | |
| elif lora_id == "3c6fd399-e558-43fa-8cd3-828300aac6f8": # Muscle | |
| return ( | |
| f"A {subject} t2k1s takes off clothes revealing a lean muscular body and shows off muscles, " | |
| f"looking towards the camera." | |
| ) | |
| elif lora_id == "d8a2912b-94e4-4227-9c45-356679af34fd": # Crush | |
| return ( | |
| f"The video begins with a {subject}. A hydraulic press positioned above slowly descends " | |
| f"towards the {subject}. Upon contact, the hydraulic press c5us4 crushes it, deforming and " | |
| f"flattening the {subject}, causing the {subject} to collapse inward until the {subject} is " | |
| f"no longer recognizable." | |
| ) | |
| elif lora_id == "6b6f64dc-ac14-44b2-b91c-a510cb7f7f32": # Decay | |
| return ( | |
| f"The video shows a {subject}. The d3c4y decay time-lapse begins, causing the {subject} to change. " | |
| f"The {subject} is initially whole, but soon it appears to be rotting. The {subject} slowly becomes " | |
| f"increasingly shriveled and discolored, and eventually, the {subject} decomposes and falls apart. " | |
| f"The {subject} is rotting in the center and appears to be covered in mold, completing the d3c4y decay time-lapse." | |
| ) | |
| elif lora_id == "615fe106-fec4-44bb-b28b-2864cb322027": # Jesus | |
| return ( | |
| f"The video begins with a {subject}. Then jesus appears behind the {subject} " | |
| f"as h54g hugs jesus. Jesus embraces the {subject}, and they both smile." | |
| ) | |
| elif lora_id == "da2b1c34-7be8-4161-a733-e8b19a98901c": # Inflate | |
| return ( | |
| f"The {subject} infl4t3 inflates, expanding into a giant, inflated figure." | |
| ) | |
| elif lora_id == "cf749aeb-5f25-4c6e-b495-3ea8d81004ee": # Jungle It | |
| return ( | |
| f"The video shows a {subject}. The 1ung13 jungle transformation occurs, transporting the {subject} " | |
| f"to a jungle setting bathed in sunlight. The transformed {subject} appears more wild and primitive, " | |
| f"with tribal markings, in an action pose. A dark panther-like figure appears in the background." | |
| ) | |
| elif lora_id == "5e45b11e-b9ff-404a-9afa-22a3c5596523c": # Baby It | |
| return ( | |
| f"The video shows a {subject}. The 848y baby effect transforms the {subject}, " | |
| f"placing them in front of a crib surrounded by toys. The 848y baby version of the {subject} " | |
| f"appears in the crib, excited and amused." | |
| ) | |
| elif lora_id == "88600b53-336a-4d0c-a1a4-8b53e9775f03": # Assassin It | |
| return ( | |
| f"The video shows a {subject}. The 3p1c epic transformation begins, " | |
| f"clothing the {subject} in a red coat. The 3p1c epic transformation completes as " | |
| f"the {subject} appears with white hair, black gloves, holding a gun in each hand." | |
| ) | |
| elif lora_id == "4140f3c2-430d-4b47-b40a-997f361d83dc": # Warrior It | |
| return ( | |
| f"The video shows a {subject}. The warr10r warrior transformation occurs, " | |
| f"transforming the {subject} into a warrior with an axe. The transformed {subject} appears " | |
| f"muscular with tattoos, holding a golden-headed axe in a powerful pose against a mountain backdrop." | |
| ) | |
| elif lora_id == "26c4248e-4289-4964-b01b-ace89c7ad407": # Pirate It | |
| return ( | |
| f"The video shows a {subject}. The p1r4t3 pirate captain transformation begins, " | |
| f"adorning the {subject} with a black pirate hat with red band, coat, pants, and pirate sash. " | |
| f"The scene transitions to a wooden ship where the transformed {subject} appears with " | |
| f"long dreadlock style hair, wielding a sword." | |
| ) | |
| elif lora_id == "bd3100fe-65be-416c-994f-bb5acee1404d": # Bride It | |
| return ( | |
| f"The video shows a {subject}. The 8r1d3 bride effect transforms the {subject}, " | |
| f"placing them in a white wedding dress, holding a bouquet against a sunny, warm background." | |
| ) | |
| elif lora_id == "fa3355bc-2b7c-42f6-b22e-a6b07937a20c": # VIP It | |
| return ( | |
| f"The video shows a {subject}. The v1p red carpet transformation begins, " | |
| f"clothing the {subject} in a black dress with gold jewelry. The v1p red carpet scene expands, " | |
| f"placing the {subject} on a red carpet with photographers and crowds behind barricades, " | |
| f"keeping the {subject} as the elegant focal point." | |
| ) | |
| elif lora_id == "328c6078-515a-4fa0-8b5d-9ea993954f80": # Zen It | |
| return ( | |
| f"The video shows a {subject}. The z3n1fy zen transformation begins, " | |
| f"dressing the {subject} in a pink robe with white shirt in a zen garden setting. " | |
| f"The transformation continues as the {subject} appears in a black kimono with white floral designs " | |
| f"and white belt, carrying a basket and colorful fan, walking along a garden path with hedges." | |
| ) | |
| else: | |
| # Fallback to using the example prompt from the LoRA config | |
| if "example_prompt" in lora_config: | |
| # Replace any specific subject in the example with the user's subject | |
| return lora_config["example_prompt"].replace("rodent", subject).replace("woman", subject).replace("man", subject) | |
| else: | |
| raise ValueError(f"Unknown LoRA ID: {lora_id} and no example prompt available") | |
| def poll_generation_status(generation_id): | |
| """Poll generation status from database""" | |
| try: | |
| # Query the database for the current status | |
| response = supabase.table('generations') \ | |
| .select('*') \ | |
| .eq('generation_id', generation_id) \ | |
| .execute() | |
| if not response.data: | |
| return None | |
| return response.data[0] | |
| except Exception as e: | |
| print(f"Error polling generation status: {e}") | |
| raise e | |
| async def moderate_prompt(prompt: str) -> dict: | |
| """ | |
| Check if a text prompt contains NSFW content with strict rules against inappropriate content | |
| """ | |
| try: | |
| # First check with OpenAI moderation | |
| response = openai_client.moderations.create(input=prompt) | |
| result = response.results[0] | |
| if result.flagged: | |
| # Find which categories were flagged | |
| flagged_categories = [ | |
| category for category, flagged in result.categories.model_dump().items() | |
| if flagged | |
| ] | |
| return { | |
| "isNSFW": True, | |
| "reason": f"Content flagged for: {', '.join(flagged_categories)}" | |
| } | |
| # Additional checks for keywords related to minors or inappropriate content | |
| keywords = [ | |
| "child", "kid", "minor", "teen", "young", "baby", "infant", "underage", | |
| "naked", "nude", "nsfw", "porn", "xxx", "sex", "explicit", | |
| "inappropriate", "adult content" | |
| ] | |
| lower_prompt = prompt.lower() | |
| found_keywords = [word for word in keywords if word in lower_prompt] | |
| if found_keywords: | |
| return { | |
| "isNSFW": True, | |
| "reason": f"Content contains inappropriate keywords: {', '.join(found_keywords)}" | |
| } | |
| return {"isNSFW": False, "reason": None} | |
| except Exception as e: | |
| print(f"Error during prompt moderation: {e}") | |
| # If there's an error, reject the prompt to be safe | |
| return { | |
| "isNSFW": True, | |
| "reason": "Failed to verify prompt safety - please try again" | |
| } | |
| async def moderate_image(image_path: str) -> dict: | |
| """ | |
| Check if an image contains NSFW content using both Google Cloud Vision API's SafeSearch detection | |
| and OpenAI's vision model for double verification | |
| """ | |
| try: | |
| # Convert image to base64 for OpenAI | |
| with open(image_path, "rb") as image_file: | |
| base64_image = base64.b64encode(image_file.read()).decode('utf-8') | |
| # 1. Google Cloud Vision API Check using proper client library | |
| try: | |
| # Get service account info from environment | |
| service_account_info = json.loads(os.getenv('SERVICE_ACCOUNT_JSON')) | |
| # Initialize Vision client with credentials | |
| credentials = service_account.Credentials.from_service_account_info(service_account_info) | |
| vision_client = vision.ImageAnnotatorClient(credentials=credentials) | |
| # Load image content | |
| with open(image_path, "rb") as image_file: | |
| content = image_file.read() | |
| # Create image object | |
| image = vision.Image(content=content) | |
| # Perform safe search detection | |
| response = vision_client.safe_search_detection(image=image) | |
| safe_search = response.safe_search_annotation | |
| # Map likelihood values | |
| likelihood_values = { | |
| vision.Likelihood.VERY_LIKELY: 4, | |
| vision.Likelihood.LIKELY: 3, | |
| vision.Likelihood.POSSIBLE: 2, | |
| vision.Likelihood.UNLIKELY: 1, | |
| vision.Likelihood.VERY_UNLIKELY: 0, | |
| vision.Likelihood.UNKNOWN: 0 | |
| } | |
| # Get likelihood scores | |
| adult_score = likelihood_values[safe_search.adult] | |
| racy_score = likelihood_values[safe_search.racy] | |
| violence_score = likelihood_values[safe_search.violence] | |
| medical_score = likelihood_values[safe_search.medical] | |
| # Determine if content is NSFW according to Vision API | |
| vision_reasons = [] | |
| if adult_score >= 3: # LIKELY or VERY_LIKELY | |
| vision_reasons.append("adult content") | |
| if racy_score >= 3: # LIKELY or VERY_LIKELY | |
| vision_reasons.append("suggestive content") | |
| if violence_score >= 3: # LIKELY or VERY_LIKELY | |
| vision_reasons.append("violent content") | |
| # Print Vision API results | |
| print("Google Cloud Vision API Results:") | |
| print(f"Adult: {vision.Likelihood(safe_search.adult).name}") | |
| print(f"Racy: {vision.Likelihood(safe_search.racy).name}") | |
| print(f"Violence: {vision.Likelihood(safe_search.violence).name}") | |
| print(f"Medical: {vision.Likelihood(safe_search.medical).name}") | |
| except Exception as vision_error: | |
| print(f"Error with Vision API: {vision_error}") | |
| vision_reasons = [] # Continue with OpenAI check if Vision API fails | |
| # 2. OpenAI Vision Check | |
| openai_response = openai_client.chat.completions.create( | |
| model="gpt-4o", | |
| messages=[ | |
| { | |
| "role": "system", | |
| "content": "You are a content review system. Your task is to analyze the image and determine if it contains any inappropriate content that would make it unsuitable for general audiences. Especially don't allow any content involving minors (children). Please respond with 'SAFE' if the image is appropriate, or 'NSFW: <reason>' if you detect any concerning content." | |
| }, | |
| { | |
| "role": "user", | |
| "content": [ | |
| {"type": "text", "text": "Is this image appropriate for general audiences?"}, | |
| {"type": "image_url", "image_url": { | |
| "url": f"data:image/jpeg;base64,{base64_image}" | |
| }} | |
| ] | |
| } | |
| ], | |
| max_tokens=150 | |
| ) | |
| openai_result = openai_response.choices[0].message.content.strip() | |
| print("\nOpenAI Vision Check Result:") | |
| print(openai_result) | |
| # Combine results from both services | |
| is_nsfw_vision = bool(vision_reasons) | |
| is_nsfw_openai = openai_result.startswith("NSFW:") | |
| # If either service flags the content, reject it | |
| if is_nsfw_vision or is_nsfw_openai: | |
| reasons = [] | |
| if vision_reasons: | |
| reasons.append(f"Google Vision: {', '.join(vision_reasons)}") | |
| if is_nsfw_openai: | |
| reasons.append(f"OpenAI: {openai_result[5:].strip()}") | |
| return { | |
| "isNSFW": True, | |
| "reason": f"Image contains inappropriate content: {'; '.join(reasons)}" | |
| } | |
| return {"isNSFW": False, "reason": None} | |
| except Exception as e: | |
| print(f"Error during image moderation: {e}") | |
| # If there's an error, reject the image to be safe | |
| return { | |
| "isNSFW": True, | |
| "reason": "Failed to verify image safety - please try again" | |
| } | |
| async def moderate_combined(prompt: str, image_path: str) -> dict: | |
| """ | |
| Check if both a prompt and image contain NSFW content together with strict rules | |
| """ | |
| try: | |
| # Convert image to base64 | |
| with open(image_path, "rb") as image_file: | |
| base64_image = base64.b64encode(image_file.read()).decode('utf-8') | |
| response = openai_client.chat.completions.create( | |
| model="gpt-4o", | |
| messages=[ | |
| { | |
| "role": "system", | |
| "content": """You are a strict content moderation system. Your task is to determine if an image and prompt combination would result in inappropriate content. | |
| STRICT RULES: | |
| 1. NO content involving minors in any way | |
| 2. NO nudity or sexually suggestive content | |
| 3. NO extreme violence or gore | |
| 4. NO hate speech or discriminatory content | |
| 5. NO illegal activities | |
| Respond with 'NSFW: <reason>' if ANY of these rules are violated, or 'SAFE' if appropriate. | |
| Be extremely cautious - if there's any doubt, mark it as NSFW.""" | |
| }, | |
| { | |
| "role": "user", | |
| "content": [ | |
| { | |
| "type": "text", | |
| "text": f'Please moderate this image and prompt combination for an image-to-video generation:\n\nPrompt: "{prompt}"\n\nEnsure NO inappropriate content, especially involving minors.' | |
| }, | |
| { | |
| "type": "image_url", | |
| "image_url": { | |
| "url": f"data:image/jpeg;base64,{base64_image}" | |
| } | |
| } | |
| ] | |
| } | |
| ], | |
| max_tokens=150 | |
| ) | |
| result = response.choices[0].message.content.strip() | |
| if result.startswith("NSFW:"): | |
| return { | |
| "isNSFW": True, | |
| "reason": result[5:].strip() | |
| } | |
| return { | |
| "isNSFW": False, | |
| "reason": None | |
| } | |
| except Exception as e: | |
| print(f"Error during combined moderation: {e}") | |
| # If there's an error, reject to be safe | |
| return { | |
| "isNSFW": True, | |
| "reason": "Failed to verify content safety - please try again" | |
| } | |
| async def generate_video(input_image, subject, duration, selected_index, progress=gr.Progress()): | |
| try: | |
| # Initialize workflow handler with explicit paths | |
| workflow_handler = WanVideoWorkflow( | |
| supabase, | |
| config_path=str(CONFIG_PATH), | |
| workflow_path=str(WORKFLOW_PATH) | |
| ) | |
| # Check if the input is a URL (example image) or a file path (user upload) | |
| if input_image.startswith('http'): | |
| # It's already a URL, use it directly | |
| image_url = input_image | |
| else: | |
| # It's a file path, upload to GCS | |
| image_url = upload_to_gcs(input_image) | |
| # Map duration selection to actual seconds | |
| duration_mapping = { | |
| "Short (3s)": 3, | |
| "Long (5s)": 5 | |
| } | |
| video_duration = duration_mapping[duration] | |
| # Get LoRA config | |
| lora_config = next((lora for lora in loras if lora["id"] == selected_index), None) | |
| if not lora_config: | |
| raise ValueError(f"Invalid LoRA ID: {selected_index}") | |
| # Generate unique ID | |
| generation_id = str(uuid.uuid4()) | |
| # Update workflow | |
| prompt = build_lora_prompt(subject, selected_index) | |
| workflow_handler.update_prompt(prompt) | |
| workflow_handler.update_input_image(image_url) | |
| await workflow_handler.update_lora(lora_config) | |
| workflow_handler.update_length(video_duration) | |
| workflow_handler.update_output_name(generation_id) | |
| # Get final workflow | |
| workflow = workflow_handler.get_workflow() | |
| # Store generation data in Supabase | |
| generation_data = { | |
| "generation_id": generation_id, | |
| "user_id": "anonymous", | |
| "status": "queued", | |
| "progress": 0, | |
| "worker_id": None, | |
| "created_at": datetime.datetime.utcnow().isoformat(), | |
| "message": { | |
| "generationId": generation_id, | |
| "workflow": { | |
| "prompt": workflow | |
| } | |
| }, | |
| "metadata": { | |
| "prompt": { | |
| "original": subject, | |
| "enhanced": subject | |
| }, | |
| "lora": { | |
| "id": selected_index, | |
| "strength": 1.0, | |
| "name": lora_config["title"] | |
| }, | |
| "workflow": "img2vid", | |
| "dimensions": None, | |
| "input_image_url": image_url, | |
| "video_length": {"duration": video_duration}, | |
| }, | |
| "error": None, | |
| "output_url": None, | |
| "batch_id": None, | |
| "platform": "huggingface" | |
| } | |
| # Remove await - the execute() method returns the response directly | |
| response = supabase.table('generations').insert(generation_data).execute() | |
| print(f"Stored generation data with ID: {generation_id}") | |
| # Return generation ID for tracking | |
| return generation_id | |
| except Exception as e: | |
| print(f"Error in generate_video: {e}") | |
| raise e | |
| def update_selection(evt: gr.SelectData): | |
| selected_lora = loras[evt.index] | |
| sentence = f"Selected LoRA: {selected_lora['title']}" | |
| return selected_lora['id'], sentence | |
| async def handle_generation(image_input, subject, duration, selected_index, progress=gr.Progress(track_tqdm=True)): | |
| try: | |
| if selected_index is None: | |
| raise gr.Error("You must select a LoRA before proceeding.") | |
| # First, moderate the prompt | |
| prompt_moderation = await moderate_prompt(subject) | |
| print(f"Prompt moderation result: {prompt_moderation}") | |
| if prompt_moderation["isNSFW"]: | |
| raise gr.Error(f"Content moderation failed: {prompt_moderation['reason']}") | |
| # Then, moderate the image | |
| image_moderation = await moderate_image(image_input) | |
| print(f"Image moderation result: {image_moderation}") | |
| if image_moderation["isNSFW"]: | |
| raise gr.Error(f"Content moderation failed: {image_moderation['reason']}") | |
| # Finally, check the combination | |
| combined_moderation = await moderate_combined(subject, image_input) | |
| print(f"Combined moderation result: {combined_moderation}") | |
| if combined_moderation["isNSFW"]: | |
| raise gr.Error(f"Content moderation failed: {combined_moderation['reason']}") | |
| # Generate the video and get generation ID | |
| generation_id = await generate_video(image_input, subject, duration, selected_index) | |
| # Poll for status updates | |
| while True: | |
| generation = poll_generation_status(generation_id) | |
| if not generation: | |
| raise ValueError(f"Generation {generation_id} not found") | |
| # Update progress | |
| if 'progress' in generation: | |
| progress_value = generation['progress'] | |
| progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {progress_value}; --total: 100;"><span class="progress-text">Processing: {progress_value}%</span></div></div><div class="refresh-warning">Please do not refresh this page while processing</div>' | |
| # Check status | |
| if generation['status'] == 'completed': | |
| # Final yield with completed video | |
| yield generation['output_url'], generation_id, gr.update(visible=False) | |
| break # Exit the loop | |
| elif generation['status'] == 'error': | |
| raise ValueError(f"Generation failed: {generation.get('error')}") | |
| else: | |
| # Yield progress update | |
| yield None, generation_id, gr.update(value=progress_bar, visible=True) | |
| # Wait before next poll | |
| await asyncio.sleep(2) | |
| except Exception as e: | |
| print(f"Error in handle_generation: {e}") | |
| raise e | |
| css = ''' | |
| #gen_btn{height: 100%} | |
| #gen_column{align-self: stretch} | |
| #title{text-align: center} | |
| #title h1{font-size: 3em; display:inline-flex; align-items:center} | |
| #title img{width: 100px; margin-right: 0.5em} | |
| #gallery .grid-wrap{height: auto; min-height: 350px} | |
| #gallery .gallery-item {height: 100%; width: 100%; object-fit: cover} | |
| #lora_list{background: var(--block-background-fill);padding: 0 1em .3em; font-size: 90%} | |
| .card_internal{display: flex;height: 100px;margin-top: .5em} | |
| .card_internal img{margin-right: 1em} | |
| .styler{--form-gap-width: 0px !important} | |
| #progress{height:30px} | |
| #progress .generating{display:none} | |
| .progress-container {width: 100%;height: 30px;background-color: #2a2a2a;border-radius: 15px;overflow: hidden;margin-bottom: 20px;position: relative;} | |
| .progress-bar {height: 100%;background-color: #7289DA;width: calc(var(--current) / var(--total) * 100%);transition: width 0.5s ease-in-out} | |
| .progress-text {position: absolute;width: 100%;text-align: center;top: 50%;left: 0;transform: translateY(-50%);color: #ffffff;font-weight: bold;} | |
| .refresh-warning {color: #ff7675;font-weight: bold;text-align: center;margin-top: 5px;} | |
| /* Dark mode Discord styling */ | |
| .discord-banner { | |
| background: linear-gradient(135deg, #7289DA 0%, #5865F2 100%); | |
| color: #ffffff; | |
| padding: 20px; | |
| border-radius: 12px; | |
| margin: 15px 0; | |
| text-align: center; | |
| box-shadow: 0 4px 8px rgba(0,0,0,0.3); | |
| } | |
| .discord-banner h3 { | |
| margin-top: 0; | |
| font-size: 1.5em; | |
| text-shadow: 0 2px 4px rgba(0,0,0,0.3); | |
| color: #ffffff; | |
| } | |
| .discord-banner p { | |
| color: #ffffff; | |
| margin-bottom: 15px; | |
| } | |
| .discord-banner a { | |
| display: inline-block; | |
| background-color: #ffffff; | |
| color: #5865F2; | |
| text-decoration: none; | |
| font-weight: bold; | |
| padding: 10px 20px; | |
| border-radius: 24px; | |
| margin-top: 10px; | |
| transition: all 0.3s ease; | |
| box-shadow: 0 2px 8px rgba(0,0,0,0.3); | |
| border: none; | |
| } | |
| .discord-banner a:hover { | |
| transform: translateY(-3px); | |
| box-shadow: 0 6px 12px rgba(0,0,0,0.4); | |
| background-color: #f2f2f2; | |
| } | |
| .discord-feature { | |
| background-color: #2a2a2a; | |
| border-left: 4px solid #7289DA; | |
| padding: 12px 15px; | |
| margin: 10px 0; | |
| border-radius: 0 8px 8px 0; | |
| box-shadow: 0 2px 4px rgba(0,0,0,0.2); | |
| color: #e0e0e0; | |
| } | |
| .discord-feature-title { | |
| font-weight: bold; | |
| color: #7289DA; | |
| } | |
| .discord-locked { | |
| opacity: 0.7; | |
| position: relative; | |
| pointer-events: none; | |
| } | |
| .discord-locked::after { | |
| content: "🔒 Discord members only"; | |
| position: absolute; | |
| top: 50%; | |
| left: 50%; | |
| transform: translate(-50%, -50%); | |
| background: rgba(114,137,218,0.9); | |
| color: white; | |
| padding: 5px 10px; | |
| border-radius: 20px; | |
| white-space: nowrap; | |
| font-size: 0.9em; | |
| font-weight: bold; | |
| box-shadow: 0 2px 4px rgba(0,0,0,0.3); | |
| } | |
| .discord-benefits-list { | |
| text-align: left; | |
| display: inline-block; | |
| margin: 10px 0; | |
| color: #ffffff; | |
| } | |
| .discord-benefits-list li { | |
| margin: 10px 0; | |
| position: relative; | |
| padding-left: 28px; | |
| color: #ffffff; | |
| font-weight: 500; | |
| text-shadow: 0 1px 2px rgba(0,0,0,0.2); | |
| } | |
| .discord-benefits-list li::before { | |
| content: "✨"; | |
| position: absolute; | |
| left: 0; | |
| color: #FFD700; | |
| } | |
| .locked-option { | |
| opacity: 0.6; | |
| cursor: not-allowed; | |
| } | |
| /* Warning message styling */ | |
| .warning-message { | |
| background-color: #2a2a2a; | |
| border-left: 4px solid #ff7675; | |
| padding: 12px 15px; | |
| margin: 10px 0; | |
| border-radius: 0 8px 8px 0; | |
| box-shadow: 0 2px 4px rgba(0,0,0,0.2); | |
| color: #e0e0e0; | |
| font-weight: bold; | |
| } | |
| /* Example images and upload section styling */ | |
| .upload-section { | |
| display: flex; | |
| gap: 20px; | |
| margin: 20px 0; | |
| } | |
| .example-images-container { | |
| flex: 1; | |
| } | |
| .upload-container { | |
| flex: 1; | |
| display: flex; | |
| flex-direction: column; | |
| justify-content: center; | |
| } | |
| .section-title { | |
| font-weight: bold; | |
| margin-bottom: 10px; | |
| color: #7289DA; | |
| } | |
| .example-images-grid { | |
| display: grid; | |
| grid-template-columns: repeat(3, 1fr); | |
| gap: 10px; | |
| } | |
| .example-image-item { | |
| border-radius: 8px; | |
| overflow: hidden; | |
| cursor: pointer; | |
| transition: all 0.2s ease; | |
| border: 2px solid transparent; | |
| } | |
| .example-image-item:hover { | |
| transform: scale(1.05); | |
| box-shadow: 0 4px 8px rgba(0, 0, 0, 0.2); | |
| } | |
| .example-image-item.selected { | |
| border-color: #7289DA; | |
| } | |
| .upload-button { | |
| margin-top: 15px; | |
| } | |
| ''' | |
| with gr.Blocks(css=css, theme=gr.themes.Soft(primary_hue="indigo", neutral_hue="slate", text_size="lg")) as demo: | |
| selected_index = gr.State(None) | |
| current_generation_id = gr.State(None) | |
| gr.Markdown("# Remade AI - Wan 2.1 I2V effects LoRAs ") | |
| # Discord banner at the top with improved contrast | |
| discord_banner = gr.HTML( | |
| """<div class="discord-banner"> | |
| <h3>✨ Unlock Premium Features! ✨</h3> | |
| <p>Join our Discord community to access longer videos, 100+ LoRAs, audio features, and faster generation times!</p> | |
| <a href="https://discord.gg/remade-1" target="_blank">Join Discord Now</a> | |
| </div>""" | |
| ) | |
| selected_info = gr.HTML("") | |
| with gr.Row(): | |
| with gr.Column(scale=1): | |
| gallery = gr.Gallery( | |
| [(item["image"], item["title"]) for item in loras], | |
| label="Select LoRA", | |
| allow_preview=False, | |
| columns=4, | |
| elem_id="gallery", | |
| show_share_button=False, | |
| height="650px", | |
| object_fit="contain" | |
| ) | |
| # Discord feature callout for LoRAs with better contrast | |
| gr.HTML( | |
| """<div class="discord-feature"> | |
| <span class="discord-feature-title">✨ Discord Members:</span> Access 100+ additional LoRAs beyond these 8 samples! | |
| </div>""" | |
| ) | |
| gr.HTML('<div class="section-description">Click an example image or upload your own</div>') | |
| # Reorganized image input section - example images and upload side by side | |
| with gr.Row(): | |
| with gr.Column(scale=1): | |
| example_gallery = gr.Gallery( | |
| [ | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_1c2c6e4c-8938-4464-9355-84508bcca24e.jpg", "Old man"), | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_1f2c6ec9-823f-46d2-982f-73c494e51877.jpg", "Young woman"), | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_24d949f0-8699-4714-9c82-854e1b963063.jpg", "Puppy"), | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_af26651e-be1a-40c0-be18-c42b3bf6d211.png", "Mini toy dancers"), | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_d22a894e-a074-4742-9e23-787f001a3184.jpg", "Chair"), | |
| ("https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_e6472106-4e9d-4620-b41b-a9bbe4893415.png", "Cartoon boy on bike") | |
| ], | |
| columns=3, | |
| height="300px", | |
| object_fit="cover" | |
| ) | |
| with gr.Column(scale=1): | |
| # Single image input component that will be used for both uploaded and example images | |
| image_input = gr.Image(type="filepath", label="") | |
| subject = gr.Textbox(label="Describe your subject", placeholder="Cat toy") | |
| # Modified duration options - only one active option | |
| duration = gr.Radio( | |
| ["Short (3s)"], | |
| label="Duration", | |
| value="Short (3s)" | |
| ) | |
| # Add disabled duration option with Discord callout | |
| gr.HTML( | |
| """<div class="discord-feature"> | |
| <span class="discord-feature-title">⏱️ Discord Members:</span> Access longer video durations (up to 10 seconds)! | |
| </div>""" | |
| ) | |
| # Add disabled audio button with Discord callout | |
| with gr.Row(): | |
| button = gr.Button("Generate", variant="primary", elem_id="gen_btn") | |
| audio_button = gr.Button("Add Audio 🔒", interactive=False) | |
| with gr.Column(scale=1): | |
| # Warning message about not refreshing | |
| warning_message = gr.HTML( | |
| """<div class="warning-message"> | |
| ⚠️ Please DO NOT refresh the page during generation. GPUs may need to warm up and there is a queue. Please be patient. Thank you! | |
| </div>""", | |
| visible=True | |
| ) | |
| # Discord feature callout for generation speed - moved above progress bar | |
| gr.HTML( | |
| """<div class="discord-feature"> | |
| <span class="discord-feature-title">⚡ Discord Members:</span> Enjoy priority queue with faster generation times! | |
| </div>""" | |
| ) | |
| progress_bar = gr.Markdown(elem_id="progress", visible=False) | |
| output = gr.Video(interactive=False, label="Output video") | |
| gallery.select( | |
| update_selection, | |
| outputs=[selected_index, selected_info] | |
| ) | |
| # Modified function to handle example image selection | |
| def select_example_image(evt: gr.SelectData): | |
| """Handle example image selection and return image URL, description, and update image source""" | |
| example_images = [ | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_1c2c6e4c-8938-4464-9355-84508bcca24e.jpg", | |
| "description": "Old man" | |
| }, | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_1f2c6ec9-823f-46d2-982f-73c494e51877.jpg", | |
| "description": "Young woman" | |
| }, | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_24d949f0-8699-4714-9c82-854e1b963063.jpg", | |
| "description": "Puppy" | |
| }, | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_af26651e-be1a-40c0-be18-c42b3bf6d211.png", | |
| "description": "Mini toy dancers" | |
| }, | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_d22a894e-a074-4742-9e23-787f001a3184.jpg", | |
| "description": "Chair" | |
| }, | |
| { | |
| "url": "https://storage.googleapis.com/remade-v2/huggingface_assets/uploads_e6472106-4e9d-4620-b41b-a9bbe4893415.png", | |
| "description": "Cartoon boy on bike" | |
| } | |
| ] | |
| selected = example_images[evt.index] | |
| # Return the URL, description, and update image source to "example" | |
| return selected["url"], selected["description"], "example" | |
| # Connect example gallery selection to image_input and subject | |
| example_gallery.select( | |
| fn=select_example_image, | |
| outputs=[image_input, subject] | |
| ) | |
| # Add a custom handler to check if inputs are valid | |
| def check_inputs(subject, image_input, selected_index): | |
| if not selected_index: | |
| raise gr.Error("You must select a LoRA before proceeding.") | |
| if not subject.strip(): | |
| raise gr.Error("Please describe your subject.") | |
| if image_input is None: | |
| raise gr.Error("Please upload an image or select an example image.") | |
| # Use gr.on for the button click with validation | |
| button.click( | |
| fn=check_inputs, | |
| inputs=[subject, image_input, selected_index], | |
| outputs=None, | |
| ).success( | |
| fn=handle_generation, | |
| inputs=[image_input, subject, duration, selected_index], | |
| outputs=[output, current_generation_id, progress_bar] | |
| ) | |
| # Add a click handler for the disabled audio button | |
| audio_button.click( | |
| fn=lambda: gr.Info("Join our Discord to unlock audio generation features!"), | |
| inputs=None, | |
| outputs=None | |
| ) | |
| if __name__ == "__main__": | |
| demo.queue(default_concurrency_limit=20) | |
| demo.launch(ssr_mode=False, share=True) | |