Daniel's VTuber LoRAs
FAQ
How to generate AI images?
- There are two methods I know of,
- locally (on own computer hardware, a graphics card with 6GB VRAM is recommended, although GTX 7XX series and 2GB is bare minimum) with an interface like Automatic1111 setup guide
- online through sites with tokens or google colab (I don't know much about these methods)
What is a LoRA and how to use it?
Advanced networks that can be easily trained
Comes in .safetensor file format
How to use LoRAs:
- Navigate to the 'Extensions' tab in the WebUI
- Move your .safetensors LoRA files into
stable-diffusion-webui/models/lora
- Press the 🎴 button under 'generate' in the WebUI
- Navigate to LoRAs and hit "refresh"
- Select the desired LoRA and adjust it's strength in the prompt eg.
<lora:asuka:0.82>
How to train LoRAs?
Following and taking advice from The Other LoRA Training Rentry and ezlora
Which model do you use to generate images?
It depends on what to generate, I've had good results with VTuber mixes like hll4-p3-ep8
For lewd, mixes of VTuber models with lewd models etc.
Some models cannot do Chibi style for example, so trying out different models can get the result you want
My LoRAs are trained with NAI, so having a model with NAI as part of the mix is a good bet for consistency.
Indie VTuber LoRAs
Anny
LoRA Download: https://files.catbox.moe/xx4r9a.safetensors
Helpful tags:
Negatives:
Notes
- to remove choker, try using negative prompts collar, choker, and positive tags collarbone see annyprompt3 below
- Lowering LoRA weight to 0.7-0.8 can help with changing default outfits, while maintaining facial and hair features
- "navel" can help show the black bikini outfit, vs "navel" in negative prompt for pink outfit, more testing needed
- "fox tail" may show up even with negative prompt
AnnyPrompt1 AnnyPrompt2 AnnyPrompt3
Artemis of the Blue
LoRA Download: https://files.catbox.moe/83cvld.safetensors
Helpful tags:
Negatives: white background, simple background
Notes
- There are issues with white backgrounds showing up, along with the bangs of hair meshing with the white hoodie, it might be because too many images with a white background in training data
- There is RNG to shark fin on top of her head, it's not an ahoge so it can get it wrong. Maybe "shark girl" can help? although it can bring out shark tails and other features
ArtemisPrompt1 ArtemisPrompt2 ArtemisPrompt3
Neuro-sama v2
LoRA Download: https://civitai.com/models/80754/neuro-sama-v2
Backup link: https://files.catbox.moe/fvwioa.safetensors
Helpful tags:
neko mode
Notes
- tags "button badges,bandaid on knee,uneven legwear" are unstable, dropping these tags will give better quality results (not on model though)
- some color issues with hair ribbon, may get black hair ribbons
- no frog outfit (not enough images right now)
- best results with 0.4-0.7 strength
Numi
LoRA Download: https://files.catbox.moe/gmvbvb.safetensors
Helpful tags:
Notes
- fishnets will look bad a majority of the time
- most facial features like hair/eye color, hair length, hairclips, jewelry, earrings don't need to be tagged to get good result
- interrogate tags with deepbooru for different outfits
NumiPrompt1 NumiPrompt2 NumiPrompt3
Shylily (Three Outfits)
LoRA Download: https://civitai.com/models/82222
Backup link: https://files.catbox.moe/tfl4t2.safetensors
Helpful tags:
first outfit
bikini
third outfit (one piece swimsuit)
Notes
- From testing I've done works ok with 0.8-1 Strength, lower the strength for off-model clothing.
- "cetacean tail" as tag for tail. Tails are inconsistent so expect to batch a bunch of images to get a good gen. Tails are tagged properly so it's optional.
- "heart ahoge" as tag if ahoge doesn't show up. I left it untagged to associate it with shylily, so it shows up most of the time.
- negative prompt other outfits if parts of them show up in your gens. For example, negative "skirt", "sailor collar" etc if wanting bikini outfit.
ShylilyPrompt1 ShylilyPrompt2 ShylilyPrompt3
Snuffy (Smilfy 2.0)
LoRA Download: https://files.catbox.moe/1eoh67.safetensors
Helpful tags:
Negatives:
Notes
- interrogate tags with an image of Snuffy with deepbooru for helpful tag list, default outfit has a lot of tags
- playboy bunny outfit is possible
- most of image set is from the front, so from behind and side shots may be difficult
- fishnets look bad, only way to improve is inpainting at higher quality like 768x768 or 1024x1024
- Tail will be hit or miss
SnuffyPrompt1 SnuffyPrompt2 SnuffyPrompt3 SnuffyPrompt4
IdolEN1 LoRAs
Pochi Wanmaru
LoRA Download: https://files.catbox.moe/4etgnx.safetensors
Helpful Tags:
animal hood outfit:
Negatives:
Notes
- "pochihood" is for animal hood outfit, "pochiw" is for default outfit
- very long hair/low twintails for on model look, hair goes all over the place
- most of the tags for facial features/collar are baked in and don't need to be prompted
- full body clothing will not look on model, it's inconsistent, issues with asymmetry of the thighhighs
PochiPrompt1 PochiPrompt2 PochiPrompt3
Yuko Yurei
LoRA Download: https://files.catbox.moe/omfpwl.safetensors
Helpful Tags:
Negatives:
Notes
- absurdly longhair + low twintails for lore accurate look (very random with hair going all over the place) "short hair" can get good looking results.
- negative prompts for collars, chokers, necktie, can help if trying for different outfit
- "hood down" as tag can help remove rogue hoods showing up in images for different outfit
YukoPrompt1 YukoPrompt2 YukoPrompt3
VShojo LoRAs
Henya
LoRA Download: https://files.catbox.moe/xsj2is.safetensors
Helpful Tags:
Negatives:
Notes
- The collar is VERY baked into the design, using negative weights with jacket, collar, choker, along with positive weights collarbone can help remove it
- Lowering the Lora weight to 0.7-0.8 can help with the baked in collar
HenyaPrompt1 HenyaPrompt2 HenyaPrompt3
Closing
INACTIVE. No planned updates or releases