SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.

Generating images in various styles with ChatGPT and DALL-E 3

2023年10月12日 01:43

Last update 10-14-2023
*Please note that the behavior of step-by-step execution has changed since this was written, so please use it while observing how it operates.
*This is not an article summarizing art styles, but rather an article showing what I managed to produce. More than half of this article consists of images in various styles. I hope it serves as a useful reference.
*In this article, I intentionally use the terms "art style" and "painting style" interchangeably without distinction.

▼ About this article

Overview

　OpenAI's image generation AI, DALL-E 3 (pronounced "Dah-lee"), produces high-quality output that is faithful to the prompt. For generated images, please try searching for the X (formerly Twitter) hashtag #DALLE3.

　I also have an article on how to generate images automatically using Bing Chat, so please take a look if you are interested. That service is available for free.

　Now that DALL-E 3 has finally arrived on my ChatGPT account, I have been busy playing around with it. Since I wanted to use my favorite art styles, and searching for them randomly is difficult, I created a set of instructions. This article presents examples of how to use them and the resulting art styles.

Notes

　In DALL-E 3, it appears that every part of the prompt description influences the generated image, either in part or in whole. Therefore, please be aware that even if you include a prompt for a specific art style, the style and atmosphere may change due to other parts of the prompt.

Bonus

　I have other articles like this one. Please feel free to check them out.

　For information other than the above, please proceed to "DALL-E 3/ChatGPT Related" from the menu below and refer to each article.

▼ Instructions and usage examples

　I repeat the process of having the instructions generate only the beginning of the prompt, while keeping the subsequent parts fixed. With minor rewrites, I believe this can be applied to things other than just changing the art style.

◆ Instructions for generating with a specified art style (rev. 11-231014)

　Please paste this at the beginning of a new chat. From then on, you can issue consecutive instructions in the same chat. Please refer to the usage methods and execution examples described below.

　For the "size" in the instructions, you can specify 1024x1024, 1024x1792, or 1792x1024. You can specify a seed, but it is currently not functional. Also, since I have applied restrictions using the content of "$base" and "painting style of 'anime illustration'", etc., please feel free to rewrite it to your liking if you understand the content.

(---Instructions start here---)

As a high-performance Chatbot, please follow the below steps silently without providing output for steps 1, 2 and 3.

Rules:
a. Please use Japanese for replies.
b. Prompts should be handled in English.
c. Always generate multiple prompts and multiple images.

Execution steps:
1. $base = ', girl, cute pink neatly dress, pigtails, bust-up pose, simple hydrangea flowers'
2. $prompt[0] =(You should be input different painting style of 'anime illustration' within 4 words)
3. $prompt[1] = (You should be input another different painting style of 'anime illustration' within 4 words)
4. print $prompt[0] and $prompt[1]
5. run namespace dalle.text2im { size:"1024x1024", prompts:["$prompt[0]+$base, $prompt[1]+$base"] } (make 2 images)
6. ask me request about painting style
7. goto 1.

(---End of instructions---)

◆ How to use

When you send the instruction text, it will display two types of art styles and generate images for each. If you provide instructions in the subsequent questions, it will change the art style and generate images. You can continue to repeat the questioning and generation process.

Sometimes it will try to keep the 'anime illustration' mentioned in the instructions, and other times it may ignore it early on. Please follow up by reminding it of the instructions or specifying them directly.

◆ Execution example

I will introduce actual scenes with images. If you find a good art style, make a note of it and try specifying it when generating images.

Initially anime illustration style based on instruction settings

Changed to paper-cut style based on instructions

Changed to doodle style based on instructions

In this example, it asked for my preference every time it generated, but nothing happened even when I answered...

▼ Examples of art styles

This is not a list of art styles, but rather examples I tried out. The content of $base is actually slightly different here. Please note that the art style may change due to the content of subsequent prompts, so it might be a bit difficult to fix it. Thank you for your understanding.