Welcome to SOP Guides

1. Introduction

2. Team Structure

3. Process Flow

3.1. Module Learning & Experience Design

3.2. Individual Flow Design

3.3. Narrative Design & Skills Mapping

3.4. New Assets Creation

3.5. Annex: World Building - Learning & Experience Design

3.6. Annex: World Building - Story Design

3.7. Process Flow Conclusion

4. Caveats

4. Caveats

5. LLM Prompt Engineering Techniques

5.1. Use The Latest Model

5.2. Zero-shot Prompting

5.3. Few-shot Prompting

5.4. Chain-Of-Thought Prompting

5.5. Structuring Prompts

5.6. Describing Prompts

5.7. Editing Prompts

5.8. Extending Responses

5.9. Multiple Users Collaborating

6. Text-to-Image Prompting Engineering Techniques

6.1. General Techniques

6.2. Photography

6.3. Architecture

6.4. Various Aesthetic Styles

6.5. Product & Material

6.1. General Techniques

This section is dedicated to sharing the various prompt engineering techniques for popular text-to-image generator tools such as DALL·E, Midjourney, Stable Diffusion, and Lexica.art (Powered by Stable Diffusion).

Here are general guidelines for using these tools:

• Avoid pasting the long description generated by ChatGPT directly into the tools.
• Shorten the prompts to 50 words or less and make them concise with only relevant keywords.
• Whenever possible, keywords should be focused on the straightforward description of the physical appearance, instead of abstract description or emphasis on the mood or vibe.
• Avoid describing many different objects or focuses for each environment as Midjourney would not be able to understand it. Focus on 1-3 elements and elaborate on that.
• Not necessary to be extremely descriptive all the time. Keep it vague and let the tools surprise us.

6.1.1. Common Keywords At A Glance

This is a list of common keywords applicable to image generation tools. This list is non-exhaustive and serves to provide inspiration on how to design prompts. Refer to the subsequent sections for detailed prompt engineering techniques.

6.1.2. Seed

This technique is intended for Midjourney.

If you use the same prompt repeatedly in Midjourney, you might observe fairly unpredictable outcomes. To achieve more consistent generation results, consider setting a specific seed value.

Seeds, which are numerical values, initialize generations. Midjourney will automatically create a random seed for you if one isn't provided. Each image (including every set of 4-quadrant images) has an associated seed.

Regardless of the size, the actual numbers hold no inherent meaning. They act as placeholders for specific stylistic or thematic results; numbers are convenient and plentiful, making them ideal for this purpose.

Keep in mind that Midjourney will not generate identical outcomes with the same seed, only similar ones. If you need to produce an exact replica of an image using a certain prompt and seed, you should explore Stable Diffusion.

To find the seed of a previously generated image without specifying it initially, simply react to any of your generated images with the envelope emoji: ✉️. The Midjourney bot will then provide you with the seed.

Prompt:

/imagine prompt celadon owl pitcher --seed 123

6.1.3. Image Aspect Ratio

This technique is intended for Midjourney.

Length : Height

• -ar 1:2
• -ar 2:1
• -ar 3:2
• -ar 2:3
• -ar 16:9
• -ar 9:16

On this page:

• 6.1. General Techniques
• 6.1.1. Common Keywords At A Glance
• 6.1.2. Seed
• 6.1.3. Image Aspect Ratio