import asyncio import aiofiles import asyncpraw import os import json import logging import tempfile import numpy as np import platform import shutil from pydantic import BaseModel from typing import List, Optional from fastapi import ( FastAPI, BackgroundTasks, Request, HTTPException, Body, WebSocket, WebSocketDisconnect, ) from fastapi.responses import HTMLResponse, JSONResponse from fastapi.staticfiles import StaticFiles from fastapi.templating import Jinja2Templates from moviepy.editor import ( VideoFileClip, CompositeVideoClip, TextClip, ImageClip, ColorClip, ) from moviepy.config import change_settings from PIL import Image, ImageDraw import google.generativeai as genai from google.generativeai.types import HarmBlockThreshold, HarmCategory from concurrent.futures import ProcessPoolExecutor # Import your RedDownloader (ensure it’s installed/in your PYTHONPATH) from RedDownloader import RedDownloader # Load environment variables (make sure you have a .env file with the keys) from dotenv import load_dotenv load_dotenv() # Configure logging logging.basicConfig( level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s" ) # Global list to store live log messages live_logs = [] class LiveLogHandler(logging.Handler): def __init__(self): super().__init__() def emit(self, record): log_entry = self.format(record) live_logs.append(log_entry) # Keep only the last 100 messages to avoid unbounded growth if len(live_logs) > 100: live_logs.pop(0) # Add the custom logging handler to the root logger live_log_handler = LiveLogHandler() live_log_handler.setFormatter( logging.Formatter("%(asctime)s - %(levelname)s - %(message)s") ) logging.getLogger().addHandler(live_log_handler) if platform.system() == "Windows": # Change ImageMagick binary path (adjust as needed) change_settings( { "IMAGEMAGICK_BINARY": r"C:\Program Files\ImageMagick-7.1.1-Q16-HDRI\magick.exe" } ) # Constants and paths BACKGROUND_IMAGE = "background.png" PROCESSED_VIDEOS_FILE = "processed_videos.json" MAX_VIDEO_DURATION = 60 # seconds FONT_PATH = "font/Montserrat-Black.ttf" OUTPUT_DIR = "output" # Reddit API credentials from environment CLIENT_ID = os.environ.get("REDDIT_CLIENT_ID") CLIENT_SECRET = os.environ.get("REDDIT_CLIENT_SECRET") USER_AGENT = os.environ.get("REDDIT_USER_AGENT") USERNAME = os.environ.get("REDDIT_USERNAME") PASSWORD = os.environ.get("REDDIT_PASSWORD") # Global in‐memory settings and processing state subreddits = [ "Damnthatsinteresting", "interestingasfuck", "BeAmazed", "nextfuckinglevel", ] processing_results = {} # keys: post id; values: { status, video_url, generated_title, generated_caption } class VideoItem(BaseModel): id: str title: str ups: int url: str subreddit: str duration: Optional[int] = None thumbnail: Optional[str] = None # Limit the number of workers to a reasonable number (e.g., 2 or 4) to avoid overwhelming your system. process_executor = ProcessPoolExecutor(max_workers=2) # ------------------ Helper Functions (largely re‐using your code) ------------------ async def initialize_reddit(): return asyncpraw.Reddit( client_id=CLIENT_ID, client_secret=CLIENT_SECRET, user_agent=USER_AGENT, username=USERNAME, password=PASSWORD, ) async def load_processed_videos(): if os.path.exists(PROCESSED_VIDEOS_FILE): async with aiofiles.open(PROCESSED_VIDEOS_FILE, "r") as f: try: return json.loads(await f.read()) except json.JSONDecodeError: logging.error( "Error decoding JSON in processed_videos.json. Starting with an empty list." ) return [] return [] async def save_processed_videos(processed_videos): async with aiofiles.open(PROCESSED_VIDEOS_FILE, "w") as f: await f.write(json.dumps(processed_videos)) async def get_n_comments(submission, n=3): """Gets the top n comments from a Reddit submission.""" try: comments = await submission.comments() await comments.replace_more(limit=None) top_comments = [comment.body for comment in comments] return top_comments[:n] except Exception as e: logging.error(f"Error getting comments: {e}") return [] async def fetch_trending_videos(reddit, processed_videos, subreddit_name): """Fetch trending video posts from a subreddit without downloading them.""" posts_list = [] try: subreddit = await reddit.subreddit(subreddit_name) async for post in subreddit.hot(limit=10): if post.is_video and post.id not in processed_videos: try: duration = post.media["reddit_video"]["duration"] if duration <= MAX_VIDEO_DURATION: width = post.media["reddit_video"]["width"] height = post.media["reddit_video"]["height"] # Only include posts with a vertical-ish ratio if (height / width) >= 1.6: # --- TASK 4: Skip videos with no audio --- if not post.media["reddit_video"].get("has_audio", True): logging.warning( f"Skipping post {post.id} due to no audio" ) continue posts_list.append(post) else: logging.warning( f"Skipping post {post.id} due to ratio: width {width} > height {height}" ) else: logging.warning( f"Skipping post {post.id} due to exceeding max duration." ) except KeyError: logging.warning( f"Skipping post {post.id} due to missing video dimensions/duration." ) return posts_list except Exception as e: logging.error(f"Error fetching videos from subreddit {subreddit_name}: {e}") return [] async def download_video(url, filename): """ Offload the synchronous RedDownloader.Download call to a thread. """ loop = asyncio.get_running_loop() def blocking_download(): try: # This call is blocking. RedDownloader.Download(url, quality=720, output=filename.split('.')[0]) return os.path.exists(filename) except Exception as e: logging.error(f"Error downloading video: {e}") return False exists = await loop.run_in_executor(None, blocking_download) if exists: logging.info(f"Video downloaded to {filename}") return filename else: logging.error(f"Video file not found after download: {filename}") return None async def generate_title_and_caption(title, comments, first_frame, api_key): genai.configure(api_key=api_key) generation_config = genai.GenerationConfig( temperature=1, max_output_tokens=8192, response_mime_type="application/json" ) model = genai.GenerativeModel( model_name="gemini-2.0-flash", generation_config=generation_config, safety_settings={ HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_NONE, HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_NONE, HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: HarmBlockThreshold.BLOCK_NONE, HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_NONE, }, system_instruction=( "Generate a JSON object which contains a short youtube video title of max 15 words, " "and also a very interesting and informational paragraph describing the subject and any informative interesting knowledge derived from the comments, " "based on the provided video title, first frame image and comments info. Do not refer to the video or comments explicitly. " "Return a JSON object with keys 'title' and 'caption'." ), ) try: with tempfile.NamedTemporaryFile( suffix=".png", delete=False ) as temp_image_file: image = Image.fromarray(first_frame, mode="RGB") image.save(temp_image_file.name) temp_image_file.flush() image = Image.open(temp_image_file.name) parts = [f"Title: {title}\n\nComments: {comments}", image] response = await model.generate_content_async( [{"role": "user", "parts": parts}] ) image.close() temp_image_file.close() if os.path.exists(temp_image_file.name): os.remove(temp_image_file.name) try: response_json = json.loads(response.text) generated_title = response_json["title"] generated_caption = response_json["caption"] return generated_title, generated_caption except (json.JSONDecodeError, KeyError, TypeError) as json_err: logging.error( f"Error parsing JSON response: {json_err}, Response Text: {response.text}" ) return None, None except Exception as e: logging.error(f"Error generating title and caption: {e}") return None, None def round_corners(clip, radius): mask = Image.new("RGB", clip.size, 0) draw = ImageDraw.Draw(mask) draw.rounded_rectangle([(0, 0), clip.size], radius, fill=255) return clip.set_mask(ImageClip(np.array(mask), ismask=True)) def create_text_clip(text, font_area, font_size, color): width = font_area[2] - font_area[0] height = font_area[3] - font_area[1] txt_clip = TextClip( text, fontsize=font_size, color=color, font=FONT_PATH, size=(width, None), method="caption", align="center", ) if txt_clip.h > height: scale_factor = height / txt_clip.h new_font_size = int(font_size * scale_factor) txt_clip = TextClip( text, fontsize=new_font_size, color=color, font=FONT_PATH, size=(width, None), method="caption", align="center", ) vertical_offset = 8 txt_y = (height - txt_clip.h) / 4 + vertical_offset bg_clip = ColorClip(size=(width, height), color=(0, 0, 0, 0)) final_clip = CompositeVideoClip( [bg_clip, txt_clip.set_position((width / 2 - txt_clip.w / 2, txt_y))] ) return final_clip.set_position((font_area[0], font_area[1])) def calculate_text_color(frame, x, y, width, height): region = frame[y : y + height, x : x + width] if region.shape[-1] == 3: grayscale = np.dot(region[..., :3], [0.2989, 0.5870, 0.1140]) else: grayscale = region mean_brightness = np.mean(grayscale) relative_luminance = mean_brightness / 255 white_contrast = (1.0 + 0.05) / (relative_luminance + 0.05) black_contrast = (relative_luminance + 0.05) / (0.0 + 0.05) return "white" if white_contrast > black_contrast else "black" def create_watermark_clip( watermark_text, frame, x_pos, y_pos, width, height, video_duration ): optimal_color = calculate_text_color(frame, int(x_pos), int(y_pos), width, height) watermark_clip = TextClip( watermark_text, fontsize=30, color=optimal_color, font=FONT_PATH, ) watermark_width = watermark_clip.w watermark_x = (720 - watermark_width) / 2 return watermark_clip.set_position((watermark_x, 1140)).set_duration(video_duration) def generate_video( video_path, title, comments, api_key, post_id, watermark_text="@damnsointeresting" ): try: output_dir = os.path.join(OUTPUT_DIR, post_id) os.makedirs(output_dir, exist_ok=True) video = VideoFileClip(video_path) first_frame = video.get_frame(0) bg = ImageClip(BACKGROUND_IMAGE).set_duration(video.duration) audio = video.audio video_area = (40, 290, 680, 1220) video_width = video_area[2] - video_area[0] video_height = video_area[3] - video_area[1] video_ratio = video.w / video.h if video_ratio > video_width / video_height: new_width, new_height = video_width, int(video_width / video_ratio) else: new_width, new_height = int(video_height * video_ratio), video_height video = video.resize(width=new_width, height=new_height) x_pos = video_area[0] + (video_width - new_width) / 2 y_pos = video_area[1] + (video_height - new_height) / 2 video = video.set_position((x_pos, y_pos)) video = round_corners(video, 40) generated_title, generated_caption = asyncio.run( generate_title_and_caption(title, comments, first_frame, api_key) ) if not generated_title or not generated_caption: logging.error("Failed to generate title or caption. Skipping video.") return None, None title_clip = create_text_clip( generated_title, (45, 190, 675, 285), 35, "white" ).set_duration(video.duration) watermark_clip = create_watermark_clip( watermark_text, first_frame, 210, 1140, 300, 40, video.duration ) final = CompositeVideoClip( [bg, video, title_clip, watermark_clip], size=(720, 1280) ) final = final.set_duration(video.duration) if audio: final = final.set_audio(audio) output_filename = os.path.join(output_dir, f"{post_id}.mp4") final.write_videofile(output_filename, fps=30) caption_filepath = os.path.join(output_dir, f"{post_id}.txt") with open(caption_filepath, "w") as f: f.write( f"Title:\n{generated_title.strip()}\n\nCaption:\n{generated_caption.strip()}" ) return output_filename, generated_title, generated_caption except Exception as e: logging.error(f"Error processing video: {e}") return None, None, None finally: if "video" in locals(): video.close() if "final" in locals(): final.close() # ------------------ FastAPI App and Endpoints ------------------ app = FastAPI() # Mount static directories (for output videos and any static assets) if not os.path.exists(OUTPUT_DIR): os.makedirs(OUTPUT_DIR) if not os.path.exists("static"): os.makedirs("static") app.mount("/output", StaticFiles(directory=OUTPUT_DIR), name="output") app.mount("/static", StaticFiles(directory="static"), name="static") templates = Jinja2Templates(directory="templates") # Serve the index HTML page @app.get("/", response_class=HTMLResponse) async def index(request: Request): return templates.TemplateResponse("index.html", {"request": request}) @app.get("/api/videos") async def get_videos(): reddit = await initialize_reddit() processed_videos = await load_processed_videos() all_posts = [] for subreddit_name in subreddits: posts = await fetch_trending_videos(reddit, processed_videos, subreddit_name) for post in posts: try: duration = post.media["reddit_video"]["duration"] except KeyError: duration = None # Add preview_video if available from the reddit_video media. preview_video = None try: reddit_video = post.media.get("reddit_video") if reddit_video and "fallback_url" in reddit_video: preview_video = reddit_video["fallback_url"] except Exception: preview_video = None video_info = { "id": post.id, "title": post.title, "ups": post.ups, "url": post.url, "subreddit": subreddit_name, "duration": duration, "thumbnail": post.thumbnail if hasattr(post, "thumbnail") else None, "preview_video": preview_video, } all_posts.append(video_info) await reddit.close() all_posts.sort(key=lambda x: x["ups"], reverse=True) return JSONResponse(content=all_posts) @app.post("/api/process") async def process_selected_videos( background_tasks: BackgroundTasks, videos: List[VideoItem] = Body(...), ): if not videos: raise HTTPException(status_code=400, detail="No videos provided") for video_info in videos: video_id = video_info.id # now you can access attributes directly if video_id in processing_results: continue # already processing or processed processing_results[video_id] = {"status": "pending"} background_tasks.add_task(process_video_task, video_info.dict()) return JSONResponse( content={ "message": "Processing started", "video_ids": [video.id for video in videos], } ) async def process_video_task(video_info: dict): video_id = video_info.get("id") try: api_key = os.environ.get("GOOGLE_API_KEY") if not api_key: processing_results[video_id] = {"status": "error", "error": "Missing GOOGLE_API_KEY"} return reddit = await initialize_reddit() submission = await reddit.submission(id=video_id) comments = await get_n_comments(submission, 5) comments_string = "\n\n".join(comments) if comments else "No comments found." temp_video_filename = f"{video_id}.mp4" downloaded_video = await download_video(video_info.get("url"), temp_video_filename) if not downloaded_video: processing_results[video_id] = {"status": "error", "error": "Failed to download video"} await reddit.close() return # Offload heavy MoviePy processing using the process pool. loop = asyncio.get_running_loop() result = await loop.run_in_executor( process_executor, generate_video, # blocking function using MoviePy downloaded_video, video_info.get("title"), comments_string, api_key, f"{video_id}", "@damnsointeresting" ) if result and result[0]: output_file, generated_title, generated_caption = result video_url = f"/output/{video_id}/{video_id}.mp4" processing_results[video_id] = { "status": "completed", "video_url": video_url, "generated_title": generated_title, "generated_caption": generated_caption } else: processing_results[video_id] = {"status": "error", "error": "Video processing failed"} if os.path.exists(temp_video_filename): os.remove(temp_video_filename) await reddit.close() except Exception as e: processing_results[video_id] = {"status": "error", "error": str(e)} @app.get("/api/results") async def get_results(): return JSONResponse(content=processing_results) @app.get("/api/settings") async def get_settings(): return JSONResponse(content={"subreddits": subreddits}) @app.post("/api/settings") async def update_settings(data: dict): new_subs = data.get("subreddits") if not isinstance(new_subs, list): raise HTTPException(status_code=400, detail="subreddits must be a list") global subreddits subreddits = new_subs return JSONResponse( content={"message": "Settings updated", "subreddits": subreddits} ) @app.delete("/api/video/{video_id}") async def delete_video(video_id: str): video_folder = os.path.join(OUTPUT_DIR, video_id) if os.path.exists(video_folder): try: shutil.rmtree(video_folder) # Remove from our in‐memory processing_results if present. if video_id in processing_results: del processing_results[video_id] return JSONResponse(content={"message": f"Video {video_id} deleted."}) except Exception as e: raise HTTPException(status_code=500, detail=f"Error deleting video: {e}") else: raise HTTPException(status_code=404, detail="Video not found.") @app.websocket("/ws/logs") async def websocket_logs(websocket: WebSocket): await websocket.accept() last_index = 0 try: while True: # Sleep briefly to prevent busy looping await asyncio.sleep(1) # If new log messages are available, send them to the client. if last_index < len(live_logs): for log in live_logs[last_index:]: await websocket.send_text(log) last_index = len(live_logs) except WebSocketDisconnect: logging.info("Client disconnected from live logs websocket") # To run the app: # uvicorn main:app --reload