20240908v2

Ashley Graham from Resident Evil 4 (2005)

Trigger Words: ashleygraham

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

This version has somewhat alleviated the issue with bad ears (mainly through caption trick, but I cannot judge the relevance), but the problem still exists. To completely resolve it, one must start with the dataset.

Occasionally, the characters' lips may appear abnormal, which should be due to my insufficient refinement in the dataset caption, leading to the blending of poor images.

Her body proportions also have a slight issue in the full-body shot. And the tag 'brown sweater over shoulder, sleeveless' has overfitted, making it very hard to remove.

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.

You could add '3D' in the negative prompt to reduce the model's 3d style.

Adding tag such as 'realistic', 'realism' can enhance the features of the character.

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

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

20240714v1

It is recommended to add "3D" in the negative prompt to enhance the model's expressive capabilities. If not added, or if added in the positive prompt, it can make the result more closely resemble a in game style.

Looking forward to your comments and images

I have collected about 40 images, mainly screenshots from 3D models found online, and manually processed the main quality issues present in the images. The current problems with this version are as follows:

1. Due to the source of the dataset images being old and the resolution of the characters themselves in this version being not high, the generated images tend to have various drawing issues. A particularly serious issue is with the ears; if the ears in the dataset images could be repaired, the model trained might perform better.

2. The "pop star" costume of this character has been annotated rather carelessly by me, making it difficult to reproduce it through the prompt.