SYSTEM NOTICE

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

Membership Opened: Real-time Yuri Illustration Sharing

Hello, this is Rcat. I recently resumed image generation.

With that in mind, I have decided to open a membership as a new initiative for you.


What are the benefits of joining the membership?

Real-time sharing of generated images

I opened this specifically to do this!

By joining the exclusive Discord server, generated images will be transferred in real-time using Webhooks.
My own intentions will not interfere with the transfer, so whether they are duds, miraculous shots, or have minor issues, everything will be shared in real-time.

The images posted in articles are usually ones that suit my personal taste, but with real-time sharing, you can see the ones that don't make the cut.

Access to all articles

All past articles (Yuri) posted will be viewable.
This also includes all future posts.

Why is there a fee if it's AI?

It is mainly for operating costs (electricity bills).
Please judge whether it is worth 100 yen by considering the introduction and operating costs if you were to do it yourself.

By the way, I spent 300,000 yen on my PC setup. It is easy to understand if you choose between buying a high-end PC yourself or peeking for 100 yen.

What you can do with real-time sharing

See raw results

You can see raw generation results without my intentions intervening.
To be posted on note, there are conditions such as matching my preferences, being safe for the internet, and not having major flaws, but you can see them the moment they are generated regardless of these conditions (you can see them before I do)

By the way, if I don't specify clothing, they are very rarely missing it...
Since I absolutely cannot post those on note, in a sense, they are exclusive. It happens maybe once in a thousand images, though.

See past sharing results

As long as the history remains on Discord, you can also view past generation results.

Make requests

If it doesn't deviate too much from my policies or preferences, I will add it to the prompt if you have a request for a certain type of image.
At worst, I can even generate a few individually.
Let's discuss this with the initial members as we go along.

By the way, since the current system uses text as placeholders and generates prompts randomly from several combinations, it is easy for me to accommodate requests if you want me to add specific clothing or hairstyles.

What is the frequency of real-time image generation?

Baseline

The base is a weekly cumulative total. I plan to generate
a theoretical value of about 840 images per week based on a rate of 5 images per hour. However, taking into account potential downtime due to load from upscaling or generating other images, as well as PC maintenance or reboots, I will set
a minimum baseline of 300 images per week (5 images/hour x 12 hours x 5 days)
.

If I can't reach that goal... I'm sorry, but let's think of something else. I can probably make up for it at super-high speed on another PC.

Are you doing anything special? Or are you just generating them infinitely?

Of course, I am doing something special. Specifically, I ensure it has
powerful randomness through a two-step process. I don't think anyone else is doing LLM correction, right?

In normal image generation, you "list the elements you want to draw" and then leave the rest to the seed value. With this method,
similar images are generated infinitely, so you get bored in a day.

Therefore, in my case, I use the following method.

  • Random picking of elements
    Among the elements I want to draw, I make some of them determined randomly so that a different combination is listed every time. For example,
    just 10 hairstyles and 10 outfits create 100 patterns. This alone overwhelmingly increases the variety and makes it harder to get bored.


  • Enumeration via Text Generation AI
    Starting from this time, I have introduced Text Generation AI (hereafter LLM). Depending on the settings, an LLM will
    return slightly different answers even for the same input. I'm sure many of you have experienced this, right?
    By utilizing this
    , after creating an instruction set that lists the elements of the picture in Japanese, leaving the translation and enumeration to the LLMmeans it translates with slightly different nuances each time or spontaneously adds background effects (I instruct it that it's okay to add them), so I have become able to generate illustrations I never would have thought of myself. Also, since I can easily give instructions for English expressions I don't know, the number of things I can simply draw has increased. Please see
    here for details

Example of automatic prompt creation by LLM

Sample

Image generation is a world of gacha. There are times when good things come out in succession, and there are days when everything is a dud. In particular, Yuri is a bit difficult because it requires two people to be drawn, but sometimes there is only one, or they are cut off or split into panels.

However, by upgrading to SDXL, about 1 in 3 to 5 images is something I think is good.

For example, it's like this in sequence right?

I think about 1/2 to 1/4 are pretty good, but what do you think? It feels like a few too many good ones are coming out, but they come out at this level. *Since I make the prompt (image generation instruction) change every time, there is a possibility that it won't be good if I keep pulling a combination of strange instructions.

Dud
Personally, pants are a dud
Nonsense dud
Only one person, it's a dud
Only one person, it's a dud
Rare case!

What to do after joining

Join the Discord server

After joining, please click the exclusive invitation link from the membership screen to join.

After joining, you can see the shared channel without needing any special application.

Notes

Since I don't know what kind of images will come out, the shared channel is set as an age-restricted channel.
I'm sorry, but those who do not meet the age requirement cannot join.
I assume it's fine since joining membership = paying = being an adult, but...

Disclaimer

Please note that service may be suspended without prior notice due to equipment trouble or natural disasters.
While I will strive to provide compensation, if it becomes impossible to address the situation, please cancel your membership as soon as that becomes clear.

Finally

If you have any questions before joining, please leave a comment on this article.

#Yuri
#lily
#yuri
#ImageGeneration
#Illustration


いいなと思ったら応援しよう!