Illustrious/Noobai

Version 3 is here!

• The screencap style has been improved, and now it will be better applied and integrated even with parrots that do not use anime images in their dataset. (does not apply to realistic loras)

• -rained with a new high-quality dataset and trained with a new configuration.

• Improved color, shadows, lighting, etc.

• Backgrounds improved slightly compared to V2.

• Anatomy further improved (although this still depends largely on the model used when generating).

• Works well with a weight of 1, the minimum is 0.7 and the maximum is 1.2.

1. The official style and official art tags can help improve the results and screencap style, although they are optional.

2. Recommended to use in the negative prompt: ai-generated, pale color, will help improve backgrounds, objects, characters, and color.

3. The recommended samplers are Euler a and Euler, and the schedulers: normal, simple, Kl optimal, and DDIM.

Enjoy it and have a happy New Year!

previous versions

Fine Anime Screencap version 2 is here!

This new version has an improved training, as well as a new dataset made from scratch.

• color has been improved compared to version 1

• the screencap has been improved, now the style is stronger so it will apply to most of the characters it is used on*.

*it will depend on the base model used

• can generate better backgrounds than the default backgrounds of the base model

• interprets light and shadow better

• adapts to the anime style of the character used

*only if the character's lora was trained with some screencap images

recommendations:

sampler euler a + normal, automatic, simple (recommended) or euler

use at the positive prompt: anime screencap, anime coloring, 2d, anime style (the latter are optional)

negative: 3d, limited palette, pale color

Enjoy!

Pony

Version 2 is here and it comes with these improvements!

• Improved training

• Now when generating the image it no longer generates blur as in version 1.

  

• Much easier to generate screencap style

• Improved color generation, it is not as pale as in version 1.

• For some loras of pony trained characters it uses a weight of 1.2 to 1.5, with native characters of the model it uses 1

• 

• Recommended sampler: euler a and euler

• Add to your positive prompt: source_anime, anime screencap and to your negative prompt: 3d, photorealistic, realistic, for best results.

Enjoy it!

V1

This Lora only works with PonyXL or AutismMix XL (it can also work with blends of these).

The purpose of this Lora is to generate anime-style screencap images. PonyXL is a powerful model with many styles and characters available solely through prompts, but it cannot accurately generate the screencap style. That’s why I decided to create this Lora to fulfill this purpose.

Without applied style lora

With applied style lora

During testing, I realized that the Lora works better with AutismMix XL, as it produces a much better screencap style than Pony. However, you can still achieve good results with Pony by following the following settings:

• Sampler: Euler a

• Steps: 25-30

• CFG scale: 6-8

• Clip skip: 2

• weight from 1 to 1.3

• Any resolution compatible with SDXL

Add the following to your prompt:

• For positive prompts: source_anime, anime + Trigger of LoRa

• For negative prompts: 3d, realistic, photorealistic, source_cartoon, paint, sketch

I hope you enjoy generating images, and don’t forget to share your creations!