None of the images in the previews were up-scaled, detailed, or treated beyond adjusting the sampling steps.

This model was created with the goal of providing better stability when using Animagine V4, while giving it a cute, soft style based on a dataset extracted from my model KNK SoftShading (PonyV6).

Read the version notes below

Features

• No extensive negative tags required: Works relatively well with just a few quality control tags like "low score, worst quality". Adding more may improve results.

• Improved limb generation: Produces better ??????? right out of the box.

• LoRA compatibility: It should work well with Illustrious LoRAs.

• Faithful native style representation: Thanks to the original creators, the styles were somewhat "overcooked," which helps maintain them even after fine-tuning or merges.

• User-friendly: Simple and intuitive to use, especially compared to the base model.

• Good language compatibility: The SDXL base was re-merged into the model, slightly improving natural language comprehension.

How to Use

Since this model is purely based on Animagine V4, you should follow their guidelines. However, here are some additional recommendations:

• Follow the prompt structure:
  This makes a big difference when generating characters. You can shorten the "series" weight to make it easier to change the character's style.
  Example:
  "1girl/1boy/1other, character name, from which series, everything else in any order."

• Most of the time, you’ll need "low score" in the negative prompts, but removing it can help create fun generations and simpler styles.

Why Use This Model?

You might be wondering, "Why would I use another model if I already have PonyV6 and Illustrious?"
While this model isn't as stable as Illustrious or as rich in resources as PonyV6 (LoRAs and custom models), it does offer several advantages:

• Updated character dataset: Unlike NoobAI (which is more of a "patch" over Illustrious), this model is updated from its core.

• Better understanding of certain concepts (…most of them are NSFW, but compared with illustrious, it lacks on multi characters compositions).

• Compatibility with Illustrious LoRAs.

• Produces more vibrant and dynamic images than the base Illustrious/NoobAI models, with stronger contrast—without needing to rely on v-pred models.

• Smells like apple. 🍏

Version 2

This was a little nightmare, this model is just broken internally. I had to almost re-place the Clip using a lot of LoRas and the base SDXL. The base style was made using the data set of my model KNK HelioBlend and my lora KNK SoftShading. In general performs much better than the version 1, Ani4 base, zero or opt. I'm sure of that, but isn't good as Illustrious base models are for complex compositions and multi-characters. But I feel Ani4 better in some aspects like colors, dataset, and single character compositions.

Version 1

This model was fine-tuned using a relatively small dataset to stabilize the default style and set high-quality images as the default aesthetic. While it still has issues with hands, these are significantly reduced compared to the original model.

My goal is to continue refining this model to minimize these issues without sacrificing its flexibility.

License

This model cannot be used for commercial purposes, modifications, or distribution without explicit consent. In other aspects, it follows the Animagine License.