This is VNSPMEFPH (Visual Novel Sprite Maker + Emotion Focus + Pose Helper)
This LoRA generates visual novel character sprites in half-body format with solid backgrounds.
I created this LoRA primarily to generate emotion pack for my Sillytavern cards, but it works very well for visual novels in general.
I know that many of you will recommend using the workflow provided for this purpose in Comfyui, but after trying to use it for months with my AMD graphics card, I realized that it was literally impossible. So I created this pack accordingly. Maybe it will only help a handful of people in my situation, but it's still worth a try.
Background recommendations:
- Black backgrounds work best for knee-up shots
- White backgrounds work best for chest-up shots
(Black backgrounds are easier to remove with background removal tools)
You can still request different background colors (green, red, etc.) depending on the character.
If you have the option to generate with a transparent background directly, that should work as well.
Trigger word:
V1SU4L-N0V3L, or <lora:32545e1e-30e7-41c0-a535-97f17bf4eae5:0.8>Prompt Format :
V1SU4L-N0V3L, [character description], [emotion/expression], centered, solo, half-body, [background color]Example :
V1SU4L-N0V3L, long hair, red hair, blue eyes, 1girl, smiling, centered, solo, half-body, white backgroundNote: While all listed emotions and poses have been specifically trained, you can experiment with similar expressions.
After testing, almost all emotions/poses work, so you can use many more than those listed.
For multi-word emotions, if an emotion doesn't work, use underscores instead of spaces (e.g., crying_of_joy). That usually solves the problem.
Here's the (non-exhaustive) list:
Trained Emotions:
- Admiring
- Amused
- Angry
- Annoyed
- Approval
- Aroused
- Arrogant
- Blushing
- Bored
- Breakdown
- Calm
- Confused
- Contemptuous
- Coughing
- Crazy
- Crying
- Crying of joy
- Curious
- Desire
- Determined
- Disapproval
- Disdain
- Disgust
- Disturbed
- Drunk
- Embarrassed
- Evil
- Excited
- Exhausted
- Fake smile
- Fear
- Flustered
- Furious
- Gloomy
- Gratitude
- Grief
- Happy
- Horrified
- Hypnotized
- Hypnotizing
- Inviting
- Jealous
- Joy
- Love
- Nervous
- Neutral
- Optimism
- Overwhelmed
- Panic
- Playful
- Pleading
- Poker face
- Pouting
- Proud
- Rage
- Regret
- Relief
- Remorse
- Sad
- Sadness
- Sarcastic
- Scared
- Screaming
- Shocked
- Shy
- Sick
- Sleepy
- Smile
- Smiling
- Smirk
- Smug
- Surprise
- Suspicious
- Teasing
- Thinking
- Tired
- Trauma
- Trembling
- Tyrannical
- Upset
- Whining
- Worried
- Yandere
- YellingTrained Poses (SFW):
- Hand clasped together
- Closed eyes
- Crossed arms
- Hand on head
- Hand on chin
- Hand on face
- Waving
- Waving at viewerTrained Poses (NSFW/NSFL):
- Injured
- Bandages
- Bleeding
- Dead
- Sweating
- Vomiting
- Masturbating
- Showing buttDescription
FAQ
Comments (7)
I don't quite understand how this works. We can always generate a character portrait. How does LORA help with this?
When you generate images of this style for emotions for Sillytavern, the results are often inconsistent from one image to another. The angles will not always be the same, or the graphic style may change even if you use the same checkpoint and the same loras.
By using what I have preconfigured in this loras, you will have a better chance that all the poses you make in a row will keep the same style and consistency for all your characters. This means that in the end, you will end up with a unified library using the same type of expressions and poses.
The purpose of LORAS is that everything is already defined there. I don't know if that's clear, but take a look at my example images and you'll understand better. (More specifically, look at the “waving pose” of the girl with pink hair and the boy with black hair, and you'll quickly understand what I mean.)
@D_G3N3R4T3 Okay, all attempts to generate with this Lora on this site end in error—none of the images are generated. So I'll try locally today.
I understand that this Lora will be predominantly anime-style? Or is she flexible?
@Twostep I mainly trained it with anime images.
But I imagine that if you want to change styles, It can adapt to any style, especially with a LORA style.
And yeah I don't know what causing the bug on other site
Edit : I fixed it and it seems to work. (I posted the result here for showing it)
As a fellow SillyTavern user, thank you! This looks super helpful. Can't wait to test it!
NP, im glad you enjoy it ! It really helped me a lot
This is broken, as the generation fails with this lora.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.









