app.py

import gradio as gr
import torch
from transformers import pipeline
from diffusers import StableDiffusionPipeline

ARTICLE_GENERATOR_MODEL = "gpt2"
SUMMARIZER_MODEL = "Falconsai/text_summarization"
TITLE_GENERATOR_MODEL = "czearing/article-title-generator"
IMAGE_GENERATOR_MODEL = "prompthero/openjourney-v4"

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
print(f"{DEVICE = }")

text_generator = pipeline(
    "text-generation", model=ARTICLE_GENERATOR_MODEL, device=DEVICE
)
summarizer = pipeline("summarization", model=SUMMARIZER_MODEL, device=DEVICE)
title_generator = pipeline(
    "text2text-generation",
    model=TITLE_GENERATOR_MODEL,
    device=DEVICE,
)
image_generator = StableDiffusionPipeline.from_pretrained(
    IMAGE_GENERATOR_MODEL,
    torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
)
image_generator = image_generator.to(DEVICE)


def generate_blog_post(query, article_length, title_length, summary_length):
    print("Generating article.")
    article = text_generator(query, max_length=article_length, num_return_sequences=1)[
        0
    ]["generated_text"]
    print(f"{article = }")

    print("Generating the title.")
    title = title_generator(article, max_length=title_length, num_return_sequences=1)[
        0
    ]["generated_text"]
    print(f"{title = }")

    print("Generating the summary.")
    summary = summarizer(
        article,
        max_length=summary_length,
        min_length=min(30, summary_length),
        do_sample=False,
    )[0]["summary_text"]
    print(f"{summary = }")

    print("Generating the cover image.")
    image = image_generator(
        summary, num_inference_steps=40, guidance_scale=7.5, width=512, height=512
    ).images[0]

    return title, summary, article, image


with gr.Blocks() as iface:
    gr.Markdown("# Blog Post Generator")
    gr.Markdown(
        "Enter a topic, and I'll generate a blog post with a title, cover image, and optional summary!"
    )

    with gr.Row():
        input_prompt = gr.Textbox(lines=2, placeholder="Enter your blog post topic...")

    with gr.Row():
        generate_button = gr.Button("Generate Blog Post", size="sm")

    with gr.Row():
        with gr.Column(scale=2):
            with gr.Blocks() as title_block:
                gr.Markdown("## Title")

                with gr.Accordion("Options", open=False):
                    title_length = gr.Slider(
                        minimum=10, maximum=50, value=30, step=5, label="Title Length"
                    )
                title_output = gr.Textbox(label="Title")

            with gr.Blocks() as body_block:
                gr.Markdown("## Body")

                with gr.Accordion("Options", open=False):
                    article_length = gr.Slider(
                        minimum=100,
                        maximum=1000,
                        value=500,
                        step=50,
                        label="Article Length",
                    )
                article_output = gr.Textbox(label="Article", lines=10)

        with gr.Column(scale=1):
            with gr.Blocks() as image_block:
                gr.Markdown("## Cover Image")
                image_output = gr.Image(label="Cover Image")

            with gr.Blocks() as summary_block:
                gr.Markdown("## Summary")
                with gr.Accordion("Options", open=False):
                    summary_length = gr.Slider(
                        minimum=30,
                        maximum=200,
                        value=100,
                        step=10,
                        label="Summary Length",
                    )
                summary_output = gr.Textbox(label="Summary", lines=5)

    job = generate_button.click(
        generate_blog_post,
        inputs=[
            input_prompt,
            article_length,
            title_length,
            summary_length,
        ],
        outputs=[title_output, summary_output, article_output, image_output],
    )

iface.launch()