Type | |
Stats | 1,733 4,355 23 |
Reviews | (261) |
Published | Nov 29, 2023 |
Base Model | |
Training | Steps: 8,820 Epochs: 210 |
Usage Tips | Clip Skip: 2 Strength: 0.4 |
Trigger Words | (((looking at mirror,mirror,reflection))),henshin,henshin pose,mirror on desk, |
Hash | AutoV2 4AC6C0AFEE |
Surprisingly, creating a scene with a person standing in front of a mirror seemed challenging using only Stable Diffusion, so I created a Lora for it.
The images are centered around a person looking into a mirror, making it easier to produce such compositions with this Lora. However, as multiple instances of the same person may appear in an image, there's an element of chance in generation.
To increase the likelihood of this composition, I used a large number of similar images. Still, issues like unnatural mirror frames, directions, poses, and multiple appearances of the same person are common, so consider it lucky if the reflection perfectly matches the pose. The occurrence rate varies with the character's Lora;
for instance, Asuna (SAO) appeared in about 4 out of 8 images, whereas Takina (Lycoris) appeared in only 2 out of 8.
I also used materials specific to a high school girl's room, and prompts like "makeup, adjusting clothes" might be effective. Additionally, incorporating transformation scenes with elements like "light particles" or "glowing" could also be interesting.