Instructions:

(All official model images contain complete metadata compatible with ComfyUI, with a simple and complete recipe to generate the images step-by-step, with a simple refinement system, control net functions, and IPAdapter in Bypass mode, in case the Lora user wants to shape the images with these methods)

This Lora focuses on two main functions: Transformation and making furry characters hairless while maintaining their physical forms and features (since we don't have many models with these functions that can be used to create intermediate mid-transformation states):

>Transformation: set the Lora value between 0.75 and 1.00

Most of the time, the use of transformation tags is not necessary since the Trigger Worlds were trained on the Lora itself to generate this effect, but add them if necessary.

>To make furry characters hairless: set Lora's strength to 1.00 to 1.20

For both functions, I recommend placing this set of tags in this exact order: "f0rl4zz, furless, featherless, light skin, skin, skin tail, furless tail,"

Tip:

"f0rl4zz" its the main trigger world if you want to maximize the effect.

If the character in question is still not exactly the way you want it, try changing its weight to a higher or lower value, for example: "(renamon:0.75)"

Methodology:

This is my first LoRa, but I thoroughly studied how I could generate such effects and put into practice the idea that:

Although some modern models trained on specific datasets may, over time, generate some biases that are difficult to overcome, it is possible to reinforce the priority of an AI on something it already knows but has less strength in, without necessarily negating the concepts that used to be prioritized for it, simply by giving more emphasis to "sub-concepts".

The LoRa was created with each image in the dataset being manually selected and some images edited via Affinity + AI in various models, with tags selected one by one. I hope you like it!