Variations and Fine-Tuning

In this chapter, we will delve into the specialized techniques of variations, parameters, and sizing within the context of AI-generated art. We will explore the creation and modification of images by generating multiple versions of an idea. The section on parameters and sizing will emphasize the importance of control over image attributes and quality.

We’re going to cover the following main topics:

• Variations
• Parameters and sizing
• Inpainting and outpainting in DALL-E 2

By the end of this chapter, you will gain a deep and practical understanding of key processes in crafting AI art, which will allow you creative flexibility and empower you to explore, experiment, and fine-tune your creations with DALL-E.

Variations

Variations are slightly or significantly different versions of an image that can be created by altering that image’s prompt. Creating variations of an image idea using DALL-E 3 is a powerful way to explore different perspectives of a single concept. Let’s discuss how.

Variations in DALL-E 3 offer a rich landscape for exploration and refinement, letting creators iterate on initial concepts to arrive at a wide array of visually diverse outcomes. By tweaking the initial prompt, adjusting parameters, and using the tools available, users can navigate through countless permutations of their initial idea, breathing new life and perspectives into their creations with each variant. Now that we know what variations are, we will look at the systematic approach to creating variations in DALL-E 3.

Here’s a step-by-step guide to creating variations in DALL-E 3:

1. Access the DALL-E 3 interface: Launch the DALL-E 3 interface and ensure you are logged in...

Parameters

DALL-E operates as a multimodal version of GPT-4, boasting a robust structure fortified with 12 billion parameters. This system “exchanges text for pixels,” leveraging a vast training dataset comprised of text-image pairs sourced from the internet to facilitate this interchange.

Multimodal

Multimodal in the context of DALL-E 3 refers to its ability to understand and generate content based on inputs from multiple types of data modes, particularly text and images. This is a significant aspect of its functionality and what makes it particularly powerful as an AI model.

Defining parameters in DALL-E is a pathway to creating highly personalized and unique images, lending your distinct touch to your creations. It’s an exercise in artistic detail, where your vision guides the formulation of parameters that create visual narratives tuned to your creative preferences.

One of the best ways to get what you want when working with the parameters of DALL...

Sizing

In DALL-E 3, all generated images inherently have a square format with their dimensions set to 1024 x 1024 pixels (px). However, if you include the size in your prompt, you can create an image of the sizes 1792px by 1024px or 1024px by 1792px.

Let’s consider an example.

In Figure 3.3, we use the prompt:

Create a coffee mug that says, "TODAY is the BEST day!" make it 1792px by 1024px.

Figure 3.3: Image from the prompt, Create a coffee mug that says, “TODAY is the BEST day!” make it 1792px by 1024px

Alternatively, if the sizing isn’t specified, the default will be 1024px x 1024px, as you’ll notice in the following figure.

Figure 3.4: Another iteration of the prompt in Figure 3.3

The image size used by DALL-E 3 (at the time of publishing of this book) can be used to create prints up to 20" x 11.5" or 12" x 12" while maintaining a crisp image. See...

Inpainting and outpainting in DALL-E 2

Open AI offered inpainting and outpainting as additional features for a short time on DALL-E 2. Inpainting and outpainting are two distinct processes in image processing and computer graphics that deal with the restoration or generation of image content.

Inpainting is the process of filling in missing or corrupted parts of an image with plausible data derived from the surrounding pixels. Using the inpainting feature in DALL-E 2 allowed users to repair damaged areas of an image or fill in missing parts with content generated based on the surrounding area.

Outpainting is the process of extending the boundaries of an image by generating new content around the existing canvas based on the contextual information of the current image content. This feature allowed users to extend the boundaries of an existing image, generating additional content seamlessly connected to the outer edges of the original canvas.

These features were an enormously...

Summary

In this chapter, we learned about the critical components that facilitate the optimal use of DALL-E 3: parameters, prompt engineering, and sizing. We discovered parameters and prompt engineering—sophisticated skills—where textual inputs guide the AI to hone the generation process, bringing a conceptual vision into reality. Lastly, we learned how to control image sizing. Together, these three elements form a triad that empowers users to navigate DALL-E 3 effectively.

In the next chapter, we will discuss how to create fine art prints with DALL-E 3.