Sign In
Verified:
SafeTensor
Type
LoRA
Stats
71
44
Reviews
Published
Aug 25, 2024
Base Model
Pony
Training
Steps: 1,400
Trigger Words
xinxianying
Hash
AutoV2
55F67386AE
default creator card background decoration
2025 Year of the Snake Contest Participant
sumai's Avatar
sumai

Trigger Words: xinxianying

The training of this model and the images it generates are solely for learning purposes.

You could find the example prompts from the images above.

My prompts are basically composed in the order of [character traits] + [style] + [expression] + [clothing] + [camera and action] + [background], and you can delete or modify them as needed.

Recommended weight: 1~0.6, adjust as needed until the character's appearance meets your requirements.

Upscale needed for a better performance. Upscale value recommendation is around 1.3, denoising strength is 0.2

You could add "3D" in the negative prompt to reduce the model's 3d style. If not added, it can make the result more closely resemble a in game style.

When the character's features become very flat or there is a loss of detail, consider adding "realistic" to the positive prompt and even making its weight higher to maintain the original shape and features as much as possible.

Facial distortion may easily occur in situations such as full-body shots. If there is facial distortion, consider using ADetailer for repair.

20240825v2

There is still much room for improvement.

Currently, her special costume cannot be replicated smoothly. There are also some labeling errors with other costumes.

I have retrained it about four times. The main issue is that I haven't figured out how to control the influencing factors in the dataset.

I manually annotated and corrected many errors from the previous version. I tried to label every detail in each image in the dataset. Some say that doing this might affect the model's activation ability, which might require multiple tags to activate the character. My purpose in labeling everything out is to make the character's editable parts more numerous and easier to edit. The actual effect of this in my use is also acceptable.

20240706v1


I have manually collected 102 images as a dataset, mainly consisting of game screenshots and 3D model screenshots. I manually cropped the size of the images and corrected the image quality. The reason for manually cropping the size is that I am still afraid of the distortion that might occur in the final model-generated images if I directly use unprocessed images. If you have a better method, please share it with me.

Currently, the model may have overfitting effects. It is recommended to add '3D' in the Negative prompt: If you do not add it, the style will be more inclined towards the game style.

Xin Xianying from Dynasty Warriors 9

The training of this model and the images it generates are solely for learning purposes.

Trigger Words: xinxianying

You could find the example prompts from the images above.

Recommended weight: 1~0.6, adjust as needed until the character's appearance meets your requirements.

Upscale value recommendation is around 1.5, denoising strength is 0.2

It is recommended to add '3D' to the negative prompt,

(optional ) and 'realistic' to positive prompt.

Welcome everyone to give suggestions and comments.

Looking forward to your comments and images.