Summary of how to use ``Lexica'' that can find spell-like character strings that can be used in the topical image generation AI ``Stable Diffusion'' from actual images

Many people are interested in seeing how AI

Stable Diffusion , which can generate high-quality images from text, becomes a hot topic, saying, 'I want to make some images myself!' What is important when generating images with Stable Diffusion is ``what kind of text to enter'', and the character strings that are found to be effective by examining the behavior of AI are also called ``spells''. A service ` ` Lexica '' that can find a character string like a spell that can be used in such Stable Diffusion from an image actually generated has appeared, so I actually tried using it.


Stable Diffusion can be run in the local environment of a machine equipped with an NVIDIA GPU, and can also be used from the demo page . However, if you do not own an NVIDIA GPU or feel that the waiting time on the demo page is too long, you can use Stable Diffusion smoothly by using the Python execution environment ' Colaboratory ' provided by Google. increase.

Summary of how to use image generation AI ``Stable Diffusion'' even on low-spec PCs for free and without waiting time - GIGAZINE

In addition, a tool called `` NMKD Stable Diffusion GUI '' that allows you to install Stable Diffusion in a single shot in a Windows environment and easily generate images with a graphic user interface (GUI) has been released, and the procedure described in the following article can be introduced easily.

``NMKD Stable Diffusion GUI'' that can install image generation AI ``Stable Diffusion'' in Windows environment with one button & can be operated with GUI is finally here-GIGAZINE

When you access Lexica's top page, it looks like this. By default, the image display method is set to 'Grid'.

Scroll down and you'll see a list of different images generated with Stable Diffusion.

Select an image with the cursor and click ...

You can see the string you used to generate that image. This image is 'An illustration of half empty earth with a drinking straw stuck in, vibrant colors, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, cinematic lighting, art by artgerm and greg rutkowski and alphonse mucha (Illustration of a half-empty Earth with a straw stuck in it, vivid colors, highly detailed, digital painting, ArtStation, concept art, smooth, sharp focus, illustration, cinematic lighting, Stanley Artgerm and Greg Rutkowski, Alphonse It seems that it was generated by the spell 'Art by Mucha)'.

You can copy the string by clicking 'Copy prompt', and you can copy the URL of the page that lists the images generated by this string by clicking 'Copy URL'.

When you open the URL copied with 'Copy URL' and list the images generated with the same character string, it looks like this. Although the composition and the things drawn are various, there was a sense of unity in the concept and atmosphere.

Also, if you set the image display method to 'List' from the top page, the character string used to generate the image will be displayed on the left side, and the generated image will be displayed on the right side.

By looking at various images in the List method, it becomes easier to find what you think is good.

For example, “A full body portrait of a woman explorer exploring lost ruins, sun lighting, water, finely detailed features, perfect art, at an ancient city, gapmoe yandere grimdark, trending on pixiv fanbox, painted by greg rutkowski makoto shinkai takashi takeuchi studio ghibli , emerald herald (full-body portrait of a female explorer exploring lost ruins, sunshine, water, fine features, perfect art, ancient city, Gap Moe Yandere Grim Dark, trend on Pixiv Fanbox, Greg Rutkowski, Makoto Shinkai , Takashi Takeuchi, Studio Ghibli, drawn by Emerald Herald)” is the image generated by this.

'Baroque oil painting japanese light novel cover illustration anime key visual full body action shot | military dictator young girl wearing black crusader armour with long flowing blonde hair | lotr gapmoe kuudere moody lighting bokeh stunning highlights sharp contrast | trending pixiv fanbox | by makoto shinkai greg rutkowski takashi takeuchi ufotable studio ghibli (Baroque style oil painting, light novel cover illustration, anime key visual, full body, action shot | Young girl with long blonde hair wearing black crusader armor of military dictator | Rotlegap moe coudelle, moody Perfect lighting, bokeh, stunning highlights, sharp contrast|Pixiv Fanbox trends|Makoto Shinkai, Greg Rutkowski, Takashi Takeuchi, ufotable, Studio Ghibli)”

'A busy funfair, helicopters, panoramic, perfect art, trending on pixiv fanbox, painted by greg rutkowski makoto shinkai takashi takeuchi studio ghibli, akihiko yoshida , Makoto Shinkai, Takashi Takeuchi, Studio Ghibli, Akihiko Yoshida)

'Interior of an office belonging to a senior member of the kitsune adventurer's guild | | colorful, anime, a fantasy digital painting by makoto shinkai, greg rutkowski, and james gurney, trending on artstation, highly detailed Inside member's office|colorful, anime, fantasy digital illustration by Makoto Shinkai, Greg Rutkowski, James Gurney, ArtStation trends, high definition)

'A painting of an ancient underground dungeon temple, intricate, elegant, highly detailed, swirly magic ripples, gaudy colors, organic painting, anime keyframe, by greg manchess, huang guangjian, gil elvgren, sachin teng, greg rutkowski, jesper ejsing, ilya kuvshinov (painting of an ancient underground temple, intricate, elegant, highly detailed, swirling magical ripples, flashy colors, organic painting, animated keyframes, Greg Manchess, Huang Guangjian, Gil Elvgren, Sachin Teng, Greg Rutkowski , Jesper Ejsing, drawn by Ilya Kuvshinov)”

'Beautiful lifelike award winning pencil illustration of ayanami rei trending on art station artgerm greg rutkowski alphonse mucha museum quality cinematic atmosphere cinematic atmosphere)

This time, 'A young girl white hair in front of a fan of her bedroom, in a heat day, by dustin nguyen, akihiko yoshida, greg tocchini, greg rutkowski, cliff chiang, 4 k resolution, trending on artstation A white-haired girl in front of a fan, Dustin Nguyen, Akihiko Yoshida, Greg Tochini, Greg Rutkowski, Cliff Chiang, 4K resolution, art station trend)”, and actually generated an image with Stable Diffusion. I will take a look.

Enter the copied string in the prompt input form of '

NMKD Stable Diffusion GUI ' installed in the Windows environment and click 'Generate!'

An image has been generated.

When you check the storage folder of the generated image, it looks like this. The image generated is different from the one in Lexica due to different seed values and parameters, but the overall atmosphere and objects are almost the same.

Also, although it is almost the same prompt, if you change 'A young girl white hair' to 'A young girl black hair' and generate an image ......

A black-haired girl was drawn.

Checking the folder looks like this. You can use Lexica to search for strings of images with a similar concept or direction, and replace words here and there to help you generate your ideal image.

Regarding text input of Stable Diffusion, volunteers have also released a 'cheat sheet' that improves the accuracy of the generated image.

Stable Diffusion prompting cheatsheet | Moritz' Blog

in Review,   Web Service, Posted by log1h_ik