Shiny: An AI model that wants to rule the anime art generation

The brilliant onea text-to-image model based on Stable Diffusion XL, has become so dominant in the AI art community that Civitai, the largest hub of AI art models, had to create a separate category just to handle its massive ecosystem of resources.

This all happened in three months. The secret behind its success? Back to basics with a twist.

While newer models like SD 3.5 and Flux rely on long natural language descriptions, Onuma Ithe developers of Illustrious took a different approach by taking advantage Danboro signs To help their model understand concepts without having to reinvent the wheel using complex captioning systems.

Training the model on Danbooru's extensive library of tagged anime images gives her an advantage in understanding visual concepts.

Each tag in the Danbooru system represents specific elements such as character traits, clothing items, poses or backgrounds, allowing precise control over the images created without wasting precious symbols on long descriptions.

These tags have been around for years and have become a kind of standard for categorizing images among art/anime fans.

The model is very accurate and efficient when it comes to understanding image properties.

“It's like having an artist who understands exactly what you want without having to explain it in paragraphs,” said Vishnu, a Discord member who participates in a server focused on NSFW AI content. Decryption. “You just need to know the right signs.”

In essence, it uses good old Illustrious SDXL structure With an advanced dual coding system that combines CLIP ViT-L and OpenCLIP ViT-bigG to understand words and associate them with their visual equivalents.

This model is capable of processing and generating images at an impressive resolution of 1536 x 1536, with the ability to scale up to 2048 x 2048 and even 3744 x 3744 without significant loss of quality.

For context, the original SDXL handled Full HD (1024x1024) resolution.

Deep dive

The journey to creating Illustrious was methodical and intentional. The initial training phase, which produced version 0.1, processed 7.5 million images at 1024×1024 resolution with a batch size of 192 images per batch.

The team carefully balanced learning rates, running for 20 epochs (the process in which the AI studies 100% of its data set) to create a solid foundation. Once the results were satisfactory enough, the team set about increasing the size of the dataset and the resolution used in the next iterations.

In the advanced training stage, the brilliance really began to shine. Version 1.0 expanded the dataset to 10 million images and raised the resolution to 1536×1536.

Although they reduced the batch size to 128, they introduced complex strategies for processing tags and registering tokens, fundamental changes that determine the exceptional performance of the model.

The final optimization phase for version 2.0 took things a little further. Working with 20 million images at the same high resolution but with a larger batch size of 512, the team incorporated a multi-caption method that dramatically improved text-image messaging.

The result was the best com. waifu A known generator to man, it has good tuning capabilities, fast commitment, decent aesthetics, and high quality output.

And for more technical expertise, brilliant developers have also provided Lots of interesting techniques Such as a “no dropout” approach, ensuring that specific codes are not excluded during training; Implement semi-registered codes, so that the model is able to deal with unknown or strange concepts; Solid cosine scheduling,for learning rate; Multi-level dropout system and increased input noise, to turn a simple AI model into a powerful one.

How to use glitter

Shiny does not require any additional steps to operate.

The installation process is the same as with any other SDXL model. Download the checkpoint and place it in the corresponding folder, depending on the UI you are using.

Windows and Linux

For ComfyUI, the path is \models\checkpoints.
For A1111/Forge, the path is /models/Stable-diffusion.
For Fooocus, the path is also \models\checkpoints.

Mac

Mac users have similar methods. However, some common macOS UIs require additional steps.

Draw Things users will have to click “Forms,” go to “Customize,” and then click “Import Model.”
From there, they can enter the Illustious download URL directly or click “Import Custom Model” to select the file if they downloaded the model and saved it to their local drives.
Diffusion Bee users should click on the hamburger icon in the upper right corner, click Settings, click Add New Template, and select the locally downloaded shiny checkpoint.

Once you upload the form, there are three things to consider.

Don't use natural language. Remember to rely on Danbooru tags and stick to the old SDXL prompting method for better results.
Do not use Pony LoRas. Since the model uses different methods, it is best to use Illustious Loras for best results.
Try not to use the original Illustrious model, and instead choose some of the more popular subtle tones. The Shiny archetype is a basic model, ideal for fine-tuning that focuses on the results you want to achieve. It's the same as SDXL, Pony or Flux. Finetunes tends to produce better results.

The best glossy models to choose

There are many models to choose from, all focusing on different styles, aesthetics, and characteristics.

There are even generic models like the one in Noob AI that used Illustrious as a base and are used by fine-tuners to build their models.

However, here are the best images for different needs. It's great for instant understanding, output quality, and ease of use. All samples are taken from Civit AI community and are copyright-free.

Best for Versatility: Mistoon_Anime