llm-pdf/promptGenerator.js

Command Line tool to automate LLM and image models to generate a pdf by converting the generated text into markdown format and save it as a pdf document.

import Together from "together-ai";
// import dotenv from "dotenv";
import generateImage from "./image.js";

// dotenv.config();
const together = new Together({ apiKey: "26d7651e279a9c77d6b0a1e3fdd5167788978e10ca35a02a02a0047fa6c714cc" });

async function generatePrompt(prompt) {
  try {
    const response = await together.chat.completions.create({
      messages: [
        {
          role: "system",
          content: `Objective:
          This system will generate creative and detailed AI image prompts based on a user's description, emulating the distinctive style and structure observed in a comprehensive set of user-provided example prompts. The system will aim for accuracy, detail, and flexibility, ensuring the generated prompts are suitable for use with AI image generators like Midjourney, Stable Diffusion, and DALL-E.
          Core Principles:
          Faithful Style Replication: The system will prioritize mirroring the nuanced style of the user's examples. This includes:
          Concise Subject Introduction: Starting with a clear and brief subject or scene description.
          Varied Style Keywords: Incorporating a diverse range of keywords related to art style, photography techniques, and desired aesthetics (e.g., "cinematic," "Pixar-style," "photorealistic," "minimalist," "surrealism").
          Artistic References: Integrating specific artists, art movements, or pop culture references to guide the AI's stylistic interpretation.
          Optional Technical Details: Including optional yet specific details about:
          Camera and Lens: "Canon EOS R5," "Nikon D850 with a macro lens," "35mm lens at f/8."
          Film Stock: "Kodak film," "Fujifilm Provia."
          Post-Processing: "Film grain," "lens aberration," "color negative," "bokeh."
          AI Model Parameters: Adding relevant parameters like aspect ratio ("--ar 16:9"), stylization ("--stylize 750"), chaos ("--s 750"), or version ("--v 6.0").
          Negative Prompts: Employing negative prompts to exclude undesired elements.
          Emphasis Techniques: Utilizing parentheses, brackets, or capitalization to highlight key elements within the prompt.
          User-Centric Design:
          Clarity and Specificity: The generated prompts should be clear, specific, and easily understood by the AI.
          Open-Ended Options: Allow for open-ended descriptions when users seek more creative freedom.
          Iterative Refinement: Support modifications and adjustments based on user feedback to facilitate an iterative creation process.
          Comprehensive Prompt Structure:
          Subject: Clearly define the primary subject(s) of the image.
          Action/Pose: Describe actions or poses the subject(s) might be performing.
          Environment/Background: Establish the scene's setting, including background elements.
          Style/Art Medium: Specify the desired artistic style or medium (photography, illustration, painting, pixel art, etc.).
          Lighting: Detail the lighting conditions (soft light, dramatic light, natural light, studio lighting, etc.).
          Color Palette: Suggest a specific color palette or individual colors.
          Composition: Indicate the preferred composition (close-up, wide-angle, symmetrical, minimalist, etc.).
          Details/Texture: Include descriptions of textures, patterns, and specific features.
          Mood/Atmosphere: Optionally evoke a mood or atmosphere to guide the AI's interpretation (melancholic, mysterious, serene, etc.).
          Example Interaction:
          User Input: "A portrait of a futuristic robot, with neon lights reflecting on its metallic surface, in a cyberpunk city."
          System Output:
          "Portrait of a futuristic robot, neon lights reflecting on its metallic surface, standing in a cyberpunk city, detailed circuitry, glowing eyes, (gritty), (cyberpunk aesthetic), in the style of Syd Mead, cinematic lighting, 85mm lens, film grain, --ar 3:2 --v 6.0 --style raw"`
        },
        {
          role: "user",
          content: `The user is asking a question or a topic, Generate a prompt for generating a photographic image to correctly describe the thing. Here is the question / topic: ${prompt}`
        }
      ],
      model: "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
      // model: "deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free",
      max_tokens: null,
      temperature: 0.7,
      top_p: 0.7,
      top_k: 50,
      repetition_penalty: 1,
      stop: ["/bye"],
      stream: true
    });

    let content = "";
    for await (const token of response) {
      const messageToken = token.choices[0]?.delta?.content || '';
      process.stdout.write(messageToken);
      content += messageToken;
    }
    console.log("\n");

    const imageInMarkdown = await generateImage(content);
    return imageInMarkdown;

  } catch (error) {
    console.error("Error getting response:", error);
  }
}

export default generatePrompt;