import os os.environ['CUDA_LAUNCH_BLOCKING'] = "1" from diffusers import LDMTextToImagePipeline import gradio as gr import PIL.Image import numpy as np import random import torch import subprocess from transformers import AutoModelWithLMHead, AutoModelForCausalLM, AutoTokenizer from transformers import WhisperForConditionalGeneration, WhisperConfig, WhisperProcessor import torchaudio import nltk from pydub import AudioSegment import re from datasets import load_dataset from transformers import AutoModelWithLMHead, AutoTokenizer, set_seed, pipeline import torch from transformers import GPT2Tokenizer, GPT2LMHeadModel import torch from diffusers import StableDiffusionPipeline, AutoencoderKL, UNet2DConditionModel, PNDMScheduler, DPMSolverMultistepScheduler, LMSDiscreteScheduler from transformers import CLIPTextModel, CLIPTokenizer from tqdm.auto import tqdm from torch import autocast from PIL import Image torch_device = "cuda" if torch.cuda.is_available() else "cpu" device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') def generate_lyrics(sample): model_name = "openai/whisper-tiny.en" model_config = WhisperConfig.from_pretrained(model_name) processor = WhisperProcessor.from_pretrained(model_name) asr_model = WhisperForConditionalGeneration.from_pretrained(model_name, config=model_config) asr_model.eval() input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features transcript = asr_model.generate(input_features) predicted_ids = asr_model.generate(input_features) transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True) lyrics = transcription[0] return lyrics def generate_summary(lyrics): summarizer = pipeline("summarization", model="philschmid/bart-large-cnn-samsum") summary = summarizer(lyrics) return summary def generate_prompt(summary): # model_name = "gpt2" # tokenizer = AutoTokenizer.from_pretrained(model_name) # model = AutoModelWithLMHead.from_pretrained(model_name) # Set up GPT-2 model and tokenizer model_name = 'gpt2' tokenizer = GPT2Tokenizer.from_pretrained(model_name) model = GPT2LMHeadModel.from_pretrained(model_name) # Set the device to GPU if available model = model.to(device) # Generate prompt text using GPT-2 prompt = f"Create an image that represents the feeling of '{summary}'" # Generate the image prompt input_ids = tokenizer.encode(prompt, return_tensors='pt').to(device) output = model.generate(input_ids, do_sample=True, max_length=100, temperature=0.7) prompt_text = tokenizer.decode(output[0], skip_special_tokens=True) return prompt_text def generate_image(prompt, height = 512, # default height of Stable Diffusion width = 512 , # default width of Stable Diffusion num_inference_steps = 50 , # Number of denoising steps guidance_scale = 7.5 , # Scale for classifier-free guidance generator = torch.manual_seed(32), # Seed generator to create the inital latent noise batch_size = 1,): pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16) pipe = pipe.to(torch_device) # 1. Load the autoencoder model which will be used to decode the latents into image space. vae = AutoencoderKL.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="vae") # 2. Load the tokenizer and text encoder to tokenize and encode the text. tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14") text_encoder = CLIPTextModel.from_pretrained("openai/clip-vit-large-patch14") # 3. The UNet model for generating the latents. unet = UNet2DConditionModel.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="unet") scheduler = DPMSolverMultistepScheduler.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="scheduler") vae = vae.to(torch_device) text_encoder = text_encoder.to(torch_device) unet = unet.to(torch_device) text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer([""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt") with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) latents = torch.randn((batch_size, unet.in_channels, height // 8, width // 8), generator=generator,) latents = latents.to(torch_device) scheduler.set_timesteps(num_inference_steps) latents = latents * scheduler.init_noise_sigma for t in tqdm(scheduler.timesteps): # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample # perform guidance noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 latents = scheduler.step(noise_pred, t, latents).prev_sample # scale and decode the image latents with vae latents = 1 / 0.18215 * latents with torch.no_grad(): image = vae.decode(latents).sample # DPM Solver Multistep scheduler image = (image / 2 + 0.5).clamp(0, 1) image = image.detach().cpu().permute(0, 2, 3, 1).numpy() images = (image * 255).round().astype("uint8") pil_images = [Image.fromarray(image) for image in images] f_images = pil_images return f_images def predict(lyrics, steps=100, seed=42, guidance_scale=6.0): # print(subprocess.check_output(["nvidia-smi"], stderr=subprocess.STDOUT).decode("utf8")) generator = torch.manual_seed(seed) summary_1 = generate_summary(lyrics) prompt_text_1 = generate_prompt(summary_1[0]['summary_text']) images = generate_image(prompt= prompt_text_1, generator= generator, num_inference_steps=steps, guidance_scale=guidance_scale) # images = ldm_pipeline([prompt], generator=generator, num_inference_steps=steps, eta=0.3, guidance_scale=guidance_scale)["images"] # print(subprocess.check_output(["nvidia-smi"], stderr=subprocess.STDOUT).decode("utf8")) return images[0] random_seed = random.randint(0, 2147483647) gr.Interface( predict, inputs=[ gr.inputs.Textbox(label='Text', default='a chalk pastel drawing of a llama wearing a wizard hat'), gr.inputs.Slider(1, 100, label='Inference Steps', default=50, step=1), gr.inputs.Slider(0, 2147483647, label='Seed', default=random_seed, step=1), gr.inputs.Slider(1.0, 20.0, label='Guidance Scale - how much the prompt will influence the results', default=6.0, step=0.1), ], outputs=gr.Image(shape=[256,256], type="pil", elem_id="output_image"), css="#output_image{width: 256px}", title="Cover Generator (text-to-image)", description="Application of OpenAI tools such as Whisper, ChatGPT, and DALL-E to produce covers for the given text", ).launch()