app.py

import requests
import io
from PIL import Image
import gradio as gr
import os

# Assuming you have your API tokens set in environment variables
ZEPHYR_API_TOKEN = os.getenv("HF_API_TOKEN")
SD_API_TOKEN = os.getenv("HF_API_TOKEN")

if not ZEPHYR_API_TOKEN or not SD_API_TOKEN:
    raise ValueError("API tokens not found. Please set the ZEPHYR_API_TOKEN and HF_API_TOKEN environment variables.")

ZEPHYR_API_URL = "https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-beta"
SD_API_URL = "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-xl-base-1.0"

def query_zephyr(linkedin_text):
    # Constructing a single string prompt that combines the instruction with the LinkedIn post text
    prompt = f"Prepare a prompt for Stable Diffusion for the following LinkedIn post:"

    headers = {
        "Authorization": f"Bearer {ZEPHYR_API_TOKEN}",
        "Content-Type": "application/json",
    }

    # The `inputs` field now directly contains the prompt string
    payload = {

        "role": "system", "content": prompt,
        "role": "user", "content": linkedin_text
    }

    response = requests.post(ZEPHYR_API_URL, headers=headers, json=payload)
    if response.status_code == 200:
        return response.json()
    else:
        print(response.text)  # To get more insight into what went wrong
        raise Exception(f"Failed to query Zephyr model, status code: {response.status_code}")



def generate_image_from_prompt(prompt, negative_prompt, guidance_scale, width, height, num_inference_steps):
    headers = {"Authorization": f"Bearer {SD_API_TOKEN}"}
    payload = {
        "inputs": prompt,
        "parameters": {
            "guidance_scale": guidance_scale,
            "width": width,
            "height": height,
            "num_inference_steps": num_inference_steps,
        },
    }
    if negative_prompt:  # Add negative prompt if provided
        payload["parameters"]["negative_prompt"] = negative_prompt

    response = requests.post(SD_API_URL, headers=headers, json=payload)
    image_bytes = response.content
    image = Image.open(io.BytesIO(image_bytes))
    return image

def generate_image_from_linkedin_text(linkedin_text, negative_prompt, guidance_scale, width, height, num_inference_steps):
    # Generate a prompt from the LinkedIn text using Zephyr
    zephyr_response = query_zephyr(linkedin_text)
    if zephyr_response and isinstance(zephyr_response, list):
        generated_prompt = zephyr_response[0].get("generated_text", "")
    else:
        raise ValueError("Unexpected response format from Zephyr model.")

    # Use the generated prompt to create an image with Stable Diffusion
    if generated_prompt:
        image = generate_image_from_prompt(generated_prompt, negative_prompt, guidance_scale, width, height, num_inference_steps)
        return image, generated_prompt
    else:
        raise ValueError("Failed to generate a prompt from the LinkedIn text.")

iface = gr.Interface(
    fn=generate_image_from_linkedin_text,
    inputs=[
        gr.Textbox(label="LinkedIn Message", placeholder="Enter LinkedIn message here..."),
        gr.Textbox(label="Negative Prompt", placeholder="Enter a negative prompt here (optional)..."),
        gr.Slider(label="Guidance Scale", minimum=1, maximum=20, step=0.1, value=7.5),
        gr.Slider(label="Width", minimum=768, maximum=1024, step=1, value=1024),
        gr.Slider(label="Height", minimum=768, maximum=1024, step=1, value=768),
        gr.Slider(label="Number of Inference Steps", minimum=20, maximum=50, step=1, value=30)
    ],
    outputs=[
        gr.Image(type="pil"),
        gr.Label(label="Generated Prompt")
    ],
    title="Generate Images from LinkedIn Messages",
    description="Enter a LinkedIn message to generate a creative prompt with Zephyr, which is then used to generate an image with Stable Diffusion. Image parameters can be adjusted."
)

iface.launch()