Maison AI: Generate contest images with the image generation function
This time's goal
I would like to use MaisonAI's image generation function to generate an image to enter the contest. We will also verify the functionality of additional modifications.
The contest to enter is accelerando.Ai's "Accessories that Match Brand Design" Contest . I would like to generate images of accessories that would go well with the dress below.
I decided to create an accessory that would go well with the gray outfit on the right, and before I started generating it, I gave a quick idea of what kind of accessory I wanted.
Generate accessory images with Maison AI
① Preparation for image generation
- Image type : Product photos worn by models
- Accessory type : necklace
- Color : gold
- Design : Curved lines, bold, classy design
- Lighting : Highlight the product
- Background : simple and sophisticated
② About the prompt
People who have been generating images with other generation AI tools such as Midjorney may be confused because the results will be different when using MaisonAI (StableDiffusion). As for negative prompts, I think it would be a good idea to generate them and add ones that you think you don't want to appear.
This time I tried running it with the following prompt.
Prompt used
Model wearing gold necklace, with a bold, round design, styled in a modern fashion, highlighted against a simple, sophisticated background
(Japanese: Model wearing a gold necklace, bold and rounded design, simple and sophisticated background)
Negative prompts (prompts that you do not want to appear in the results)
Watermark, No ugly shapes, avoid poorly balanced compositions, conservative, Too thin model, NSFW, Clothes that are less revealing, blurry image, white of the eye, eyes out of focus.
(Japanese: watermark logos, ugly shapes, avoid unbalanced compositions, conservative, underweight models, NSFW, skimpy clothing, blurry images, white eyes, eyes out of focus)
*Even if it is not included in the prompt, the image may become blurred (bottom of the image) when the AI determines that highly revealing clothing is being generated. That's why we include "NSFW (abbreviation for Not Safe for Work), Clothes that are less revealing, blurry image."
You can also find out more about negative prompts below.
Maison AI image generator negative prompt usage and recommended list
Leave the following default settings.
③Let's try it
Here is an image of the prompt and negative prompt prepared and executed. We have created gold accessories with a sophisticated and bold design.
④Let's make additional corrections.
I like the second image from the right, so I would like to change the design based on this one. When you hover over it, buttons for "View larger" and "Additional modifications" will appear . Click "Additional modifications" and you can add prompts for what changes you would like to make.
Instead of all the prompts you originally included, just include the prompts for the elements you want to add .
Currently, only one sheet is generated as a result of additional corrections, but it will be possible to generate multiple sheets in the future. 
Add sculptural shapes
Added "sculptural shape" as an additional modification (center of image). It has become a unique and nice shape.
I added "inspiration from plants" to the additionally corrected image. The plant now has a flowing shape. (Image right)
If you make additional corrections to something that has undergone additional corrections, it seems that it tends to take over the shape before the corrections.
Add plant images
I then went back to the originally generated image and made additional modifications. The prompt remained " inspiration from plants ." The circular earrings are shaped like a round flower, giving it a plant feel. (center of image)
When I made additional corrections to the results and generated it by inputting `` inspiration from plants '', the design turned out to be a little greener and more plant-like. I can no longer feel the impression of gold. (Image right) 
add pearl
Once again, I went back to the first generated image and added "Add a single pearl". The gold circle turned into a pearl sphere, and a single pearl was also generated on the necklace. (center of image)
Furthermore, when I added "Add a lots of pearls", the chain part of the necklace became many small pearls, and the image changed dramatically. (Image right) 
Add star image
Lastly, I added and edited the original image to say "Inspired by the shape of star" and it looked really cool, so I'm going to submit it for the contest! I tried to put it together easily. 
If you add a long prompt
I tried adding three elements, but I still couldn't express it well. The fewer additional prompts, the better. 
summary
- AI image generation is easier to work on if you consider the elements and create a prompt.
- There are two types of additional corrections: one is to perform additional corrections on an initially generated image, and the other is to perform additional corrections on an image that has been additionally corrected.
- It's best to keep your prompts simple, specific, and clear.
- It seems better to set a negative prompt.
``I've created an image with a composition that I like overall, but I want to change some parts or have a little more variation.'' In such cases, I think additional corrections are very effective.
With the additional correction function, there is no need to launch other software and make corrections in order to make partial changes, and once you get the hang of it, you will be able to create images that you like. I would be happy if you could try using this blog as a reference.
