import gradio as gr import torch import moviepy.editor as mpe from PIL import Image, ImageDraw, ImageFont from transformers import AutoTokenizer, AutoModelForSeq2SeqLM from min_dalle import MinDalle from gtts import gTTS from pydub import AudioSegment import nltk import textwrap import os import glob import subprocess import imageio_ffmpeg # Ensure 'punkt' is downloaded for nltk try: nltk.data.find('tokenizers/punkt') except LookupError: nltk.download('punkt') # Download FFmpeg using imageio_ffmpeg (more robust) try: imageio_ffmpeg.get_ffmpeg_exe() print("FFmpeg downloaded successfully (if not already present).") except Exception as e: print(f"Error downloading FFmpeg using imageio_ffmpeg: {e}") raise description = "Video Story Generator with Audio \n PS: Generation of video by using Artificial Intelligence by dalle-mini and distilbart and gtss " title = "Video Story Generator with Audio by using dalle-mini and distilbart and gtss " tokenizer = AutoTokenizer.from_pretrained("sshleifer/distilbart-cnn-12-6") model = AutoModelForSeq2SeqLM.from_pretrained("sshleifer/distilbart-cnn-12-6") device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model.to(device) print(device) def get_output_video(text): inputs = tokenizer(text, max_length=1024, truncation=True, return_tensors="pt").to(device) summary_ids = model.generate(inputs["input_ids"]) summary = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False) plot = list(summary[0].split('.')) ''' The required models will be downloaded to models_root if they are not already there. Set the dtype to torch.float16 to save GPU memory. If you have an Ampere architecture GPU you can use torch.bfloat16. Set the device to either "cuda" or "cpu". Once everything has finished initializing, float32 is faster than float16 but uses more GPU memory. ''' def generate_image( is_mega: bool, text: str, seed: int, grid_size: int, top_k: int, image_path: str, models_root: str, fp16: bool, ): model = MinDalle( is_mega=is_mega, models_root=models_root, is_reusable=True, is_verbose=True, dtype=torch.float16 if fp16 else torch.float32, device=device ) image = model.generate_image( text, seed, grid_size, top_k=top_k, is_verbose=True ) return image generated_images = [] for senten in plot[:-1]: image = generate_image( is_mega=True, text=senten, seed=1, grid_size=1, # param {type:"integer"} top_k=256, # param {type:"integer"} image_path='generated', models_root='pretrained', fp16=True, ) generated_images.append(image) # Step 4- Creation of the subtitles sentences = plot[:-1] num_sentences = len(sentences) assert len(generated_images) == len(sentences), print('Something is wrong') # We can generate our list of subtitles from nltk import tokenize c = 0 sub_names = [] for k in range(len(generated_images)): subtitles = tokenize.sent_tokenize(sentences[k]) sub_names.append(subtitles) # Step 5- Adding Subtitles to the Images def draw_multiple_line_text(image, text, font, text_color, text_start_height): draw = ImageDraw.Draw(image) image_width, image_height = image.size y_text = text_start_height lines = textwrap.wrap(text, width=40) for line in lines: line_width, line_height = font.getbbox(line)[2:4] # Use getbbox for better size calculation draw.text(((image_width - line_width) / 2, y_text), line, font=font, fill=text_color) y_text += line_height def add_text_to_img(text1, image_input): ''' Testing draw_multiple_line_text ''' image = image_input fontsize = 20 # Increased font size path_font = "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf" if not os.path.exists(path_font): # Try alternative location on different systems path_font = "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf" if not os.path.exists(path_font): print("Font file not found. Subtitles might not be rendered correctly.") path_font = None if path_font is not None: try: font = ImageFont.truetype(path_font, fontsize) text_color = (255, 255, 0) text_start_height = 200 draw_multiple_line_text(image, text1, font, text_color, text_start_height) except Exception as e: print(f"Error loading or using font: {e}") return image generated_images_sub = [] for k in range(len(generated_images)): imagenes = generated_images[k].copy() text_to_add = sub_names[k][0] result = add_text_to_img(text_to_add, imagenes) generated_images_sub.append(result) # Step 7 - Creation of audio c = 0 mp3_names = [] mp3_lengths = [] for k in range(len(generated_images)): text_to_add = sub_names[k][0] print(text_to_add) f_name = 'audio_' + str(c) + '.mp3' mp3_names.append(f_name) # The text that you want to convert to audio mytext = text_to_add # Language in which you want to convert language = 'en' # Passing the text and language to the engine, # here we have marked slow=False. Which tells # the module that the converted audio should # have a high speed myobj = gTTS(text=mytext, lang=language, slow=False) # Saving the converted audio in a mp3 file named sound_file = f_name myobj.save(sound_file) audio = AudioSegment.from_file(sound_file, format="mp3") duration = len(audio) / 1000 mp3_lengths.append(duration) print(duration) c += 1 # Step 8 - Merge audio files cwd = os.getcwd().replace(chr(92), '/') export_path = 'result.mp3' silence = AudioSegment.silent(duration=500) full_audio = AudioSegment.empty() for n, mp3_file in enumerate(mp3_names): mp3_file = mp3_file.replace(chr(92), '/') print(n, mp3_file) # Load the current mp3 into `audio_segment` audio_segment = AudioSegment.from_mp3(mp3_file) # Just accumulate the new `audio_segment` + `silence` full_audio += audio_segment + silence print('Merging ', n) # The loop will exit once all files in the list have been used # Then export full_audio.export(export_path, format='mp3') print('\ndone!') # Step 9 - Creation of the video with adjusted times of the sound c = 0 file_names = [] for img in generated_images_sub: f_name = 'img_' + str(c) + '.jpg' file_names.append(f_name) img.save(f_name) c += 1 print(file_names) clips = [] d = 0 for m in file_names: duration = mp3_lengths[d] print(d, duration) clips.append(mpe.ImageClip(m).set_duration(duration + 0.5)) d += 1 concat_clip = mpe.concatenate_videoclips(clips, method="compose") concat_clip.write_videofile("result_new.mp4", fps=24) # Step 10 - Merge Video + Audio movie_name = 'result_new.mp4' export_path = 'result.mp3' movie_final = 'result_final.mp4' def combine_audio(vidname, audname, outname, fps=24): my_clip = mpe.VideoFileClip(vidname) audio_background = mpe.AudioFileClip(audname) final_clip = my_clip.set_audio(audio_background) final_clip.write_videofile(outname, fps=fps) combine_audio(movie_name, export_path, movie_final) # create a new file # Cleanup intermediate files for f in file_names: os.remove(f) for f in mp3_names: os.remove(f) os.remove("result_new.mp4") os.remove("result.mp3") return 'result_final.mp4' text = 'Once, there was a girl called Laura who went to the supermarket to buy the ingredients to make a cake. Because today is her birthday and her friends come to her house and help her to prepare the cake.' demo = gr.Blocks() with demo: gr.Markdown("# Video Generator from stories with Artificial Intelligence") gr.Markdown( "A story can be input by user. The story is summarized using DistillBART model. Then, then it is generated the images by using Dalle-mini and created the subtitles and audio gtts. These are generated as a video.") with gr.Row(): # Left column (inputs) with gr.Column(): input_start_text = gr.Textbox(value=text, label="Type your story here, for now a sample story is added already!") with gr.Row(): button_gen_video = gr.Button("Generate Video") # Right column (outputs) with gr.Column(): output_interpolation = gr.Video(label="Generated Video") gr.Markdown("

Future Works

") gr.Markdown( "This program text-to-video AI software generating videos from any prompt! AI software to build an art gallery. The future version will use Dalle-2 For more info visit [ruslanmv.com](https://ruslanmv.com/) ") button_gen_video.click(fn=get_output_video, inputs=input_start_text, outputs=output_interpolation) demo.launch(debug=False)