Noisy_Element (SD 1.5 LCM) and VAA/VALite (SD 1.5)
Noisy_Element
This version is almost like VALite, if VALite was strictly an LCM checkpoint, or used VAA v1.0 instead of VAA v2.0.
This is essentially an LCM version of Victorian Anime Art, with some slight differences.
For one, this was made using mostly VAA v1.0 stuff, as opposed to version 2.0 for Victorian Anime Art. As a result of that, it's not quite up to par as far as the Victorian Anime theme I was going for.
This model has a VAE already baked into it and doesn't require an external LCM LoRA to get good results.
Noisy_Element Recommended usage
Set the sampler to LCM, decrease the CFG to no more than about 2, and reduce the steps to around 8. Enter prompt as normal and generate as normal. I've noted that in hires fix, a denoising strength of 0.7 can often mess things up. If it happens to you, I recommend lowering the denoising strength to about 0.4-0.5.
It has been noted that it can work on other samplers and higher CFG, but it tends to give weird artifacts in the generated images. Best bet is to increase the steps to around 30+ and hope something good comes out, if you plan to try other samplers.
I tried to train this model to make good hands, but being an SD1.5 model, hands are still a mixed bag at best. So, if you know a good method to get consistent hands, I recommend using that method here too. Same thing with feet.
Have fun~
Note:
In case you were wondering, Noisy Element = Monet Einsley. Anagrams are fun, but I probably should feel insulted by that lol.
Victorian Anime Art SD1.5
VAA
To be blunt, this model is essentially 50% Victorian Oil Painting, and 50% anime. It could go either way depending on the prompt. No VAE required, compatible with LCM (if you include the LCM LoRA).
VALite
This version was intended to be more anime and less oil painting. If I had to put a ratio on it, I'd say it was 75% anime, 25% oil painting. But I believe it could go either way, depending on the prompt, and if you incorporate the VAA v2.0 LoRA with it. No VAE needed, and also LCM compatible if you use the appropriate LoRA and settings.
Summary for VAA/VALite
The goal of this model is to produce art with a Victorian oil-painting mixed with anime aesthetic. VAA and VALite were trained on the VAA v2.0 standard at varying degrees, with VALite being more anime and less oil painting. Further training might be necessary to go heavy into the Victorian oil painting aesthetic.
All models have a VAE already baked in, and are compatible with LCM samplers--provided you also use an LCM LoRA. Like with Noisy_Element, I attempted to train these models to get decent hands, but it's a mixed bag at best, I find.
Recommended usage
Non-LCM samplers: Do as you normally do with standard SD.15 checkpoints, and it should work.
LCM samplers: apply in the appropriate LCM LoRA, set the CFG to 1-2, lower steps to about 6-10. If using ADetailer, be sure to also apply the LCM LoRA in the prompt, if you're using a different prompt.
Version differences
My intent was for VAA to be more Victorian oil painting and less anime, but at times, it seems like it's 50-50; my intent for VALite was to be more anime and less Victorian oil painting, but it feels like it's 75-25. So, from my perspective, the differences seem a bit subtle, depending on what you're doing. To be fair though, I have yet to invoke the prompt "Victorian oil painting", but I have invoked "Mona Lisa" sometimes and have been surprised with the results. Also trying prompts like "Monet" (the painter), "Da Vinci", and other such Renaissance/Victorian era artists might yield some surprising results. Best to experiment and see what works for you. If you want more Victorian Oil painting as much as I do, you could try using the Victorian Anime Art LoRAs and see if that helps; they're good but somewhat flawed, so I wouldn't recommend using weights more than 0.7 on them. It's probably high time for me to develop a better VAA v3.0 standard. If I do that, I'll try to make a more oil-painting based model with that and upload it here.
Best of luck to you~
Note:
Yes, I'm aware that the Mona Lisa is in fact Rennaissance, and not Victorian, but I wanted to show Mona wearing modern clothing so that's what I went with lol
Example Prompts
Noisy_Element (non-LCM sampler)
(Masterpiece, best quality, ultra-detailed, art by Da Vinci, oil painting:1.4)
BREAK (transparent Egyptian gown with gold lining, gold bra, Usekh Collar, cleavage:1.3)
BREAK (platinum blonde hair, massive breasts:1.3), half updo with spherical bun, wavy hair with blunt bangs, floral hair accessories
BREAK portrait of a girl in a cafe, city street, blue eyes, glossy lipstick, gorgeous smile
Negative prompt: (worst quality, lowres, bad anatomy:1.4) bad_prompt_version2-neg badhandv4 NegfeetV2 ng_deepnegative_v1_75t
Steps: 70, Size: 512x512, Seed: 3794655636, Model: Noisy_Element, Version: v1.7.0, Sampler: DPM++ 3M SDE Exponential, CFG scale: 7, Clip skip: 2, NegfeetV2: df90b1ff666d, Pad conds: True, badhandv4: 5e40d722fc3d, Model hash: 3e87df8658, Hires steps: 20, oil painting: 1.4)\nBREAK blonde hair, blue eyes, blunt bangs, gorgeous smile\nBREAK glossy red lipstick, long eyelashes, blush", Hires upscale: 2, Hires upscaler: 4x-AnimeSharp, ADetailer model: face_yolov8m.pt, ADetailer steps: 28, ADetailer prompt: "(Masterpiece, best quality, ultra-detailed, art by Monet, ADetailer version: 24.1.2, Denoising strength: 0.45, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ng_deepnegative_v1_75t: 54e7e4826d53", ADetailer inpaint width: 1024, "bad_prompt_version2-neg: 6f35e7dd816a, ADetailer inpaint height: 1024, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer use separate steps: True, ADetailer inpaint only masked: True, ADetailer use inpaint width height: True
Noisy_Element (LCM sampler)
Mona Lisa but with short (light green hair and eyes, bobcut:1.4) with blunt bangs, medium breasts
BREAK wearing a silk white kimono with floral patterns, a single white flower in her hair
BREAK Renaissance oil painting style, highly detailed scenery, mountains in the distance
BREAK art by Monet
Negative prompt: (worst quality, bad quality, bad anatomy, frame, borders, text, watermark:1.45)
Steps: 8, Size: 512x512, Seed: 3185870127, Model: Noisy_Element, Version: v1.7.0, Sampler: LCM, CFG scale: 1.5, Clip skip: 2, Pad conds: True, Model hash: 3e87df8658, green eyes: 1.3), bobcut with blunt bangs, medium breasts\nBREAK wearing a silk white kimono with camellia patterns on it, a single white flower in her hair\nBREAK Renaissance oil painting style\nBREAK long eyelashes, mascara, eyeliner, eyeshadow, lip gloss\nBREAK gorgeous smile, blushing, soft lighting\nBREAK art by Monet", Hires upscale: 2, Hires upscaler: 4x-UltraSharp, ADetailer version: 24.1.1, Denoising strength: 0.5, ADetailer model 4th: face_yolov8m.pt, ADetailer steps 4th: 16, ADetailer prompt 4th: "Mona Lisa with (short green hair, ADetailer y offset 4th: 20, ADetailer mask blur 4th: 8, ADetailer confidence 4th: 0.3, ADetailer dilate erode 4th: 60, ADetailer inpaint padding 4th: 32, ADetailer denoising strength 4th: 0.5, ADetailer use separate steps 4th: True, ADetailer inpaint only masked 4th: True
VAA (+ LCM LoRA and LCM sampler)
(best quality, highest quality, HDR, raytracing, 8k:1.4)
BREAK (bob cut with blunt bangs, light green, green eyes:1.4)
BREAK (full body shot of a woman standing in a vast palace hallway, gold lining, windows and doors, gold furniture, shiny floors:1.4)
BREAK (wearing a blue strapless Egyptian robe, bare legs, pelvic curtain, gold anklet, gold bracelet, gold necklace, golden sandals, embroidery:1.4) <lora:PAseer-SD15-LCM Quick:1>
Negative prompt: (worst quality, bad quality, bad anatomy, 6finger, 6toes:1.4) bad_prompt_version2-neg badhandv4 EasyNegative NegfeetV2 ng_deepnegative_v1_75t
Steps: 16, 8k: 1.4)\nBREAK (hand focus[, pink fingernails, Size: 512x768, Seed: 3485407319, Model: VAA, Sampler: LCM, CFG scale: 1.5, Clip skip: 2, Model hash: 1f08026b7d, Denoising strength: 0.5, ADetailer model 2nd: hand_yolov8s.pt, ADetailer model 3rd: foot_yolov8x_v2.pt, ADetailer steps 2nd: 8, ADetailer prompt 2nd: "(high resolution photograph, best quality, masterpiece, highest quality, HDR, ADetailer prompt 3rd: "(high resolution photograph, best quality, masterpiece, highest quality, HDR, <PAseer-SD15-LCM Quick:1>", ADetailer mask blur 2nd: 4, ADetailer confidence 2nd: 0.42, ADetailer dilate erode 2nd: 4, ADetailer inpaint width 2nd: 1024, ADetailer inpaint height 2nd: 1024, ADetailer inpaint padding 2nd: 100, ADetailer denoising strength 2nd: 0.4, ADetailer use separate steps 2nd: True, ADetailer inpaint only masked 2nd: True, ADetailer use inpaint width height 2nd: True
VAA (non-LCM sampler)
(masterpiece, best quality, Victorian oil painting style, Renaissance oil painting style, art by Monet:1.4)
BREAK full body shot of Mona Lisa with sideswept bangs, green eyes, gorgeous smile, large breasts, freckles
BREAK standing in a lush forest, dappled sunlight, wearing a frilly dress, Mary Janes with ruffled socks
Negative prompt: (bad quality, worst quality, bad anatomy, frame, borders:1.4) bad_prompt_version2-neg badhandv4 NegfeetV2 ng_deepnegative_v1_75t EasyNegative
Steps: 45, Seed: 2447904672, CFG scale: 7
VALite (+LCM LoRA and LCM sampler)
(best quality, highest quality, HDR, raytracing, 8k:1.4)
BREAK <lora:ishtartest:1> (Ishtar, red eyes, blushing:1.4)
BREAK (full body shot of a woman standing on a beach, crashing waves, rocks, vegetation, palm trees, parasol, benches, towels, island resort:1.4)
BREAK (wearing a striped bikini, barefoot:1.4) <lora:PAseer-SD15-LCM Quick:1>
Negative prompt: (worst quality, bad quality, bad anatomy, 6finger, 6toes:1.4) bad_prompt_version2-neg badhandv4 EasyNegative NegfeetV2 ng_deepnegative_v1_75t
Steps: 16, 8k: 1.4)\nBREAK (hand focus[, pink fingernails, Size: 512x768, Seed: 1977774679, Model: VALite, Sampler: LCM, CFG scale: 1.5, Clip skip: 2, Model hash: 60cc489e7c, Denoising strength: 0.5, ADetailer model 2nd: hand_yolov8s.pt, ADetailer model 3rd: foot_yolov8x_v2.pt, ADetailer steps 2nd: 8, ADetailer prompt 2nd: "(high resolution photograph, best quality, masterpiece, highest quality, HDR, ADetailer prompt 3rd: "(high resolution photograph, best quality, masterpiece, highest quality, HDR, <PAseer-SD15-LCM Quick:1>", ADetailer mask blur 2nd: 4, ADetailer confidence 2nd: 0.42, ADetailer dilate erode 2nd: 4, ADetailer inpaint width 2nd: 1024, ADetailer inpaint height 2nd: 1024, ADetailer inpaint padding 2nd: 100, ADetailer denoising strength 2nd: 0.4, ADetailer use separate steps 2nd: True, ADetailer inpaint only masked 2nd: True, ADetailer use inpaint width height 2nd: True
VALite (non-LCM sampler)
(masterpiece, highest quality, best quality, UHD, HDR, raytracing, 8k:1.4)
BREAK (full body:1.4) shot of an elegant woman with massive breasts wearing a (semitransparent silk dress with lace hemline, strapless, sleeveless, very short skirt, barefoot:1.3), cleavage
BREAK short brown hair, green eyes, freckles
BREAK standing in front of a garden next to a (house made of glass and gold:1.3), mountain, valley, fjord
Negative prompt: (worst quality, bad quality, lowres, bad anatomy, mutation, 6fingers, 6toes:1.45) bad_prompt_version2-neg badhandv4 EasyNegative NegfeetV2 ng_deepnegative_v1_75t
Steps: 30, 8k: 1.4)\nBREAK (short brown hair, green eyes, floral hair ornaments, Size: 512x512, Seed: 318375158, Model: VALite, blush: 1.2)\nBREAK beautiful eyes, long eyelashes", Sampler: DPM++ 3M SDE Exponential, CFG scale: 5, Clip skip: 2, freckles: 1.4)\nBREAK (gorgeous smile, gorgeous face, Model hash: 60cc489e7c, ADetailer model: face_yolov8m.pt, ADetailer steps: 40, ADetailer prompt: "(masterpiece, highest quality, best quality, UHD, HDR, raytracing, Denoising strength: 0.7, ADetailer mask blur: 4, ADetailer model 2nd: foot_yolov8x.pt, ADetailer confidence: 0.3, ADetailer prompt 2nd: "(masterpiece, highest quality, best quality, UHD, HDR, raytracing, ADetailer dilate erode: 20, ADetailer inpaint width: 1024, ADetailer inpaint height: 1024, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.5, ADetailer use separate steps: True, ADetailer inpaint only masked: True, ADetailer use inpaint width height: True
Additional Resources
https://huggingface.co/MonetEinsley/ADetailer_CM
This is a foot model for Adetailer
https://huggingface.co/MonetEinsley/Victorian_Anime_Art
This is the VAA 1.0/2.0 LoRAs. They're overfit somewhat, so using a weight of no more than 0.7 is ideal if you plan to use it.