
Last-minute entry for the Halloween contest. If you like the model, I'd really appreciate if you like this resource so I can compete w/the mummy girlies ?
As Halloween approaches, we imagine monsters lurking in shadows and ghosts walking among us. Yet sometimes, the worst monsters aren't the ones that hide under our beds or in dark corners – they're the ones that live within us.
A doppelganger (German for "double-walker") is a person's exact double. This concept captures Halloween's eeriness. What could be more unsettling than meeting another "you"? Someone who knows ALL your memories and darkest thoughts you've never told anyone? I'm sure I’m not the only one who gets freaked out imagining the reflection in the mirror just… blinking on its own.
This is a multi-concept LoRA, and honestly, it’s the toughest one I've made because of how SDXL handles mirrors and reflections (or rather, doesn’t handle them at all). But I managed to put something together that works pretty consistently (by SDXL standards, anyway). It responds better to anime models since it was trained entirely on anime images, though it can work with western styles too. Just don’t expect it to do realism. Results will vary depending on how well the base mode/LoRA's that you are using knows the concept of "mirror".
3 STAGES:
The model is divided between 3 stages/concepts:
-
Stage 1: The character is looking at the viewer, not at the mirror.
hauntedmirror, stage-1, reflection, dark persona, dual persona, different reflection, looking at viewer, mirror, upper body, symmetry, gothic dress, black hair, blue eyes, black eyes, empty eyes, yandere, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
-
Stage 2: The character faces the mirror, looking directly at their reflection.
hauntedmirror, stage-2, reflection, dark persona, dual persona, different reflection, looking at another, eye contact, mirror, upper body, gothic dress, black hair, blue eyes, black eyes, empty eyes, yandere, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
-
Stage 3: The character’s face isn’t visible, only their reflection is, revealing the “evil twin” version of themselves.
hauntedmirror, stage-3, 1girl, solo, reflection, looking at mirror, mirror, from behind, upper body, gothic dress, black hair, blue eyes, black eyes, empty eyes, yandere, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
Activation tags:
You will ALWAYS need to use these:
hauntedmirror, reflection, mirror, black eyes, empty eyes, yandere
Stage 1 tags:
-
stage-1: Start with this one.
-
different reflection + dual persona + dark persona: Follow with these.
-
looking at viewer: Pretty self explanatory.
-
looking back: Character's back will be facing the viewer.
-
front view: Character's body will be facing the viewer.
-
(blue) (purple) (red) eyes: Specify the character's eye color to distinguish it from the reflection.
hauntedmirror, stage-1, reflection, different reflection, dual persona, dark persona, looking at viewer, mirror, front view, full body, full-length mirror, wariza, holding stuffed toy, gothic dress, black hair, blue eyes, black eyes, empty eyes, hand on mirror, yandere, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
Stage 2 tags:
-
stage-2: Start with this one.
-
different reflection + dual persona + dark persona: Follow with these.
-
looking at mirror: Follow with this one.
-
looking at another: Follow with this one too.
-
eye contact: Character will make eye contact with the reflection.
-
looking down/up: Add "looking down/up" if needed.
-
from side + profile: The scene is viewed from its side.
-
front view: Character's body will be facing the viewer.
-
from behind: The scene is viewed from behind
-
faceless: Use to hide the character's face; it can help with consistency.
-
(blue) (purple) (red) eyes: Specify the character's eye color to distinguish it from the reflection. Do not use if using the "faceless" tag.
hauntedmirror, stage-2, mirror, reflection, different reflection, dual persona, dark persona, looking at another, looking down, from side, profile, hand in mirror, holding hands, blue eyes, gothic dress, black hair, black eyes, empty eyes, hand on mirror, yandere, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
Stage 3 Tags:
-
stage-3: Start with this one.
-
1girl + solo: Use it so only one face is visible
-
looking at mirror: Straightforward.
-
from behind: Shows the character from a back perspective. You can only use this perspective in stage 3.
hauntedmirror, reflection, mirror, black eyes, empty eyes, yandere, 1girl, solo, looking at mirror, from behind, gothic dress, masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest
General Tags:
Here are some tags that you can use to improve your generations ^-^
full-length mirror + full body | For a full body shot.
upper body | If you prefer just the upper body.
cowboy shot | A mid-length shot, not quite full body.
from above | Should work fine for that perspective.
dutch angle | Adds a dramatic touch.
hand on mirror + hand up + holding hands | Just remember that hands aren't there yet.
symmetry | Can be hit or miss, but worth trying.
close-up | Different results depending on the stage.
shaded face | To make the reflection more eerie.
matching outfit | For similar but not identical outfits.
crack | If you want the mirror to be broken
@ @ | For a stylized eye effect.
heads together | I recommended ControlNet instead of relying on this tag.
bathroom + sink| Most of the images used for training were in bathrooms, so it’s well recognized by the model.
ALL THE EXAMPLE IMAGES WERE GENERATED WITH Illustrious! It should work the same for Pony though. For pony-specific examples, please refer to the showcase images. Those were generated with AutismMix-Confetti.
Honestly, I don’t think the example images do the model justice but I’m not the best at prompting (which is why I’m in the training category instead of image creation, lol). But if you get any cool pictures, definitely share them here! It’ll help others generate awesome images too, especially since this model can be tricky to work with. I'll be tipping the most creative images.
Recommended weight: 0.7 to 1 (Increase the weight if you are having trouble with the mirror.)
Enjoy the spooky season! ?
描述:
训练词语: hauntedmirror
名称: haunted_mirror_illustrous.safetensors
大小 (KB): 223102
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success