Zsófia a Sárkány - Zhophia the Dragoness
Details
Download Files
About this version
Model description
[Zsófia a Sárkány – Zhophia the Dragoness
](https://www.facebook.com/zsofia.sarkany.16/)
Before discovering Stable Diffusion, I once drew an image of a green dragon girl with a red hair which kinda gave this Hungarian flag like look to her. I used this single drawing for a one-image training LoRA model, and after many prompt testing and creating possible variations, I must say this LoRA model I trained here is quite satisfactory for many reason:
I. Design Accuracy:
The original character, is this dataset’s dataset only protagonist, is a Hungarian Green Scaled Dragon with a Long Red hair, two horns, and blue eyes. Her name is Zsófia, which is pronounced ‘Zhophia’ in English.
To get her original design here are following trigger word and reinforcement tags you should add to your prompts in order to get her generated:
The main trigger word at the float value of 1.0 is ‘Zhophia’
Reinforcement tags: ‘dragoness, green skin, two horns, red hair, blue eyes’
Zhophia, dragoness, green skin, two horns, red hair, blue eyes, … your prompt
II. LoRA compatiblity:
It is very compatible with other LoRA models, especially goes well with concept LoRA models. For example, if you want her to be muscular or whatever body type preference you fancy the most, then she will adopt that body type while maintaining her original look.
**III. Friendly to Adding more features:
**If you prefer dragon girls with wings, it can be done through positive prompts; all you need to is add the words in double brackets prompt,….((dragon wings, closed wings)), … prompt.
IV. Design Flexibility:
By design flexibility what I actual mean is that this LoRA is very flexible to create other characters. With this LoRA you can create your own scalie character based on your preference, that character will pick up Zhophia’s artstyle.
You can also alter the species, in this case, I did a crocodile and a kobold, as well as a black wingless dragon bully.
The main trigger word at float value of 1.0 is ‘Zhophia’
Reinforcement tags: ‘NONE!
Example: Zhophia, species (e.g. crocodile), color skin, color scales, other features (e.g. black hair, orange highlights), ….
V: Float Strength 'Slide-able'
Lora:name:#\
The # - is the float value, which is a decimal number. 1.0 is 100% is you want closer style and details to Zsofia (Zhophia) art style, but if you have a LoRA with a particular style and make a unique art style mixture or closer to your original prompt style, I recommend the float value strength (i.e. # ) to be no lower than 0.55 (55%):
_If you like to experiment with this lora as a slider, then I truly recommend you to acquire echoAlpha V5.5 if you seek for more aesthetic results. You can find the link in the suggested resources on this lora pa_ge
Additional Note: The trigger word 'Zhophia' was replaced with ((elephant)) and all other reinforcement tags related to features of Zhophia's appearance were removed as well. In addition to that, the following images also has another lora included, which is one of my conceptual type lora used to generate preferable muscular anatomy and the art-style with it too. So the conceptual lora was kept at 100%; while the Zsofia lora was used as a demonstration as a slider to see if you can use this lora as an style mixture:
Examples:
1: # = 1.0 (100%)
2: # = 0.55 (55%)
3: # = 0 (i.e. lora is offline, lol)
V. Possible Nuisance_
You might encounter multicolor imbalances; for example, her torso is light green but her breast are somehow dark green. It is rare occurrences, though._
For best results, I truly recommend using the checkpoint model called EchoAlpha V5.5. I personally think it is so far the best checkpoint for SDXL series I have ever experiences. They are very good at handling character models.
Join our YIFF_AI Discord server for more information and updates: