Beyond Human-Designed Prompts with Generative Ideation

Generative ideation is defined as the generation of ideas and content without human intervention. Developing the ideation automation process remains within the scope of human designers and developers. In production, an application will not require human intervention with prompts.

A generative ideation ecosystem automates the production from an idea to text and image content. The development phase requires highly skilled human AI experts. For an end user, the ecosystem is a click-and-run experience.

In Part I, we will first define the new world of generative ideation ecosystems. You will learn to think of yourself as a music composer and the AI tools as instruments. In music, the toolbox includes instruments such as guitars, drums, saxophones, and keyboards. In this chapter, you will compose generative ideation with a fantastic orchestra that contains ChatGPT/GPT-4, Llama 2, Midjourney, Stable Diffusion, and Microsoft Designer...

Part I: Defining generative ideation

Ideation refers to creating new ideas through intuitive thinking, brainstorming, and other imaginative methods.

Generative ideation leverages Generative AI to create content such as text, text-to-image, image-to-text, and image-to-image artifacts.

Automated generative ideation takes generative ideation a step further and suppresses the need for human-designed prompts. The goal of this chapter is to show how to build such tools.

Let’s define an automated ideation architecture.

Automated ideation architecture

Figure 20.1 shows the framework implemented in this chapter:

Figure 20.1: Generative AI ideation framework

In this chapter, we will cover the four-step development process of automated generative ideation:

1. No prompt

   Automated features, such as Python code, will automate document retrieval, parsing, and search in a now classical Generative AI Retrieval Augmented Generation (RAG) process.

   ...

Part II: Automating prompt design for generative image design

For an AI developer, going beyond human-designed prompts with Generative AI is not very different from creating classical complex reports based on database queries:

• The end user often doesn’t want to have anything to do with making the report.
• The end user defines the report.
• A developer creates SQL queries, for example.
• The developer creates a click-and-run interface.
• The end user works on the automatically generated report.

Why does this work? The key concept is a closed environment. In a small company or a small department of a large corporation, the tasks can be well defined, the data well identified, and the users willing to automate work that takes them hours to do and infringes on their free time.

In this section, we will:

• Work in a closed environment.
• Reproduce classical software automation for end users.
• Build a use case for a marketing...

Part III: Automated generative ideation with Stable Diffusion

This section demonstrates how to automate the process of instructing a Large Language Model to generate prompts automatically without human intervention, as shown in Figure 20.11. We will illustrate the process with the example of a small business that doesn’t have marketing resources but would like to generate images for posters.

Figure 20.11: A fully automated Generative AI ideation process

Figure 20.11 shows the fully automated process:

1. No prompt: The prompt will not be interactive. The program will create the prompt automatically. As such, AF1 is the function that takes the output of the automated instruction and sends it to an LLM as input to generate prompts automatically.
2. Generative AI (prompt generation): ChatGPT, GPT-4, will generate text-to-image prompts automatically. The AF2 function will automatically chain GPT-4 to Stable Diffusion without human intervention.
3. Generative...

The future is yours!

In this chapter, we implemented several methods to automate generative ideation. You can take this much further by imagining many other pipelines for each component of the automated pipeline:

1. No prompt, automated instructions

   This chapter illustrated one way of automating human prompts. However, each project has its constraints. Among other automated and RAG approaches, you could explore:

   • Chapter 11, Leveraging LLM Embeddings as an Alternative to Fine-Tuning. The Transfer_Learning_with_Ada_Embeddings.ipynb notebook shows how to process larger documents.
   • Chapter 15, Guarding the Giants: Mitigating Risks in Large Language Models. The Mitigating_Generative_AI.ipynb notebook shows how to use a knowledge base to retrieve information.

2. Generative AI prompts without human intervention

   In this section, we automated prompt generation with ChatGPT/GPT-4. However, you could also use other LLMs, such as Llama, as shown...

Summary

In this chapter, we first defined generative ideation. The core concept resides in one sentence:

AI has gone from executing tasks based on human prompts to automating ideation enhancing human thinking and imagination in an ethical ecosystem.

Generative ideation can be constructive and implemented without destroying human jobs. For example, generative ideation can help small businesses with no marketing resources to access human-level AI services to compete with larger entities.

We then demonstrated that generative text-to-image ideation was possible with ChatGPT/GPT-4, Llama 2, Midjourney, and Microsoft Designer.

Finally, we automated the generative text-to-image ideation pipeline with automated prompt design, ChatGPT/GPT-4, and Stable Diffusion. We saw that you could use the multiple tools explored in this book to build an effective ecosystem.

We also peeked into the future of automation and AI productivity.

You can leverage automated human thinking...

Questions

1. Generative ideation is impossible. (True/False)
2. Text-to-image technology will soon be abandoned. (True/False)
3. ChatGPT can generate prompts. (True/False)
4. Llama 2 can create text content. (True/False)
5. Midjourney is an image-to-text system. (True/False)
6. Microsoft Designer can automate ideation with the right prompt. (True/False)
7. Some companies do not have marketing resources. (True/False)
8. Stable Diffusion can help small businesses. (True/False)
9. Ethical AI can boost a career. (True/False)
10. Society can benefit from automated ideation. (True/False)

References

• Meta’s Llama 2 on Hugging Face: https://huggingface.co/meta-llama
• Touvron et al., 2023, Llama 2: Open Foundation and Fine-Tuned Chat Models: https://arxiv.org/abs/2307.09288
• Midjourney (getting started): https://www.midjourney.com/
• Discord (getting started): https://support.discord.com/hc/en-us/articles/360033931551-Getting-Started
• Discord API tutorial for Midjourney: https://medium.com/@neonforge/how-to-create-a-discord-bot-to-download-midjourney-images-automatically-python-step-by-step-guide-3e76d3282871
• Microsoft Designer: https://designer.microsoft.com/
• Immersed: https://www.immersed.com
• Pico 4: https://business.picoxr.com/

Further reading

• Hugging Face blog on Llama: https://huggingface.co/blog/llama2

Join our community on Discord

Join our community’s Discord space for discussions with the authors and other readers:

https://www.packt.link/Transformers