My English is not good, so I used ChatGPT to translate this introduction.

If you can use the Pony model, I recommend using the Pony version of the LoRA. Click here

I am a beginner at training LoRA models, so this LoRA model is a test version.

This LoRA is not only suitable for sleeping scenarios; it is a facial expression concept that can be used for any scene you imagine. It mainly affects the character's facial features, making the eyebrows and eyes appear particularly relaxed, weak, blank, and lifeless.

• Trigger words:

  HJG, empty eyes

• Trigger words when lying down to sleep(Commonly used):

  HJG, empty eyes, sleeping

• Alternative trigger words:

  HJG, empty eyes, half-closed eye

• Optional prompt words↓

drooling

open mouth

(one eye closed:1.1)

looking up

tongue

tongue out

• Negative prompt words:

  (blush:1.2), bags under eyes, lips, (eye_contact:1.2), short fingers

• It is recommended to always enable "ADetailer" to assist in fixing facial details.

• It is recommended to always enable the "Hires. fix" high-resolution repair function to ensure there are enough pixels to carve eye details.

SD1.5 performs poorly when generating inverted or side-view faces. Please use ControlNet to avoid this issue.

I recommend using another author's LoRA [eye opening (eye size) control] to fine-tune the eyes.

——————————

Reference image ①: <lora- SD1.5 v0.2-000006>, HJG, empty eyes, half-closed eye

Different checkpoints may vary, but a relatively safe range for the LoRA weight values is 0.9 - 1.2

Reference image ②: <lora- SD1.5 v0.2-000006:1.01>, HJG, empty eyes, half-closed eye

The weight value for "half-closed eye" can vary between 1.0 - 1.4 to fine-tune the degree of eye openness.

Reference Image ③: <lora- SD1.5 v0.2-000006:1.01>, HJG, empty eyes, sleeping

Adjusting the weight of "sleeping" can also fine-tune the character's eye openness, and it will likely make the character look elsewhere. My typical weight adjustment range is between 1.0 and 1.2

When using this LoRA, I really enjoy pairing it with ControlNet OpenPose. It helps maintain the correct head orientation for the character. Here, I'm sharing some preset images.

When using OpenPose, don't stick with the default parameter settings. If you're unsure what to use, you can refer to my settings:

• Control Weight: 0.6

• Starting Control Step: 0

• Ending Control Step: 0.3 - 0.7

Regarding the LoRA Block Weight plugin, I'm not sure why, but version 1.1 of this LoRA significantly weakens when using expression-type preset layering parameters. I have tried many parameter variations without much success, so I have not found suitable lbw parameters for this LoRA (‾◡◝)...

These descriptions are just part of my understanding of this LoRA model. Feel free to explore more ways to use it boldly and share with everyone! ❤️