2024.05.06 : v6a, v6b
2024.01.20 : v5c
This model was created after a brief hiatus.
It was developed with the mindset of starting anew. It may feel quite different from the existing models up to v4.
Please understand that I have only performed a simple merge.
I am adhering to a very restrictive copyright policy due to copyright issues with the merged model.
Your understanding is appreciated.
I used extension BMAB
https://github.com/portu-sim/sd-webui-bmab?tab=readme-ov-file
Sampler : DPM++ SDE Karras, DPM++ 2M Karras, DPM++ 2S a
Step : 20~25
cfg : 7.5 ~ 12.5
05-14 : v4.0
This model has tried to change so that both Westerners and Asians can express it well.
In addition, we tried to make dark expression more active.
As a result, the overall tone of the person comes out dark, but it looks better on a dark background.
* The example photos I upload are just examples, and SD environments are all different, so even if you use the same prompt, you may get different results.
Please see the image as an example only. *
Make good use of light-related prompts.
05-01 : v3.0
In version v2.3, the anime tendency was slightly visible, so I tried to make it more realistic again.
04-24 : v2.3
This version has changed mixing ratios compared to the previous v2.2 and includes some experimental attempts. As always, extensive testing has not been conducted.
I uploaded the pruned version separately, but it is currently downloading only the full version maybe there is a problem with Civitai. I will solve the problem and upload the pruned version within the 25th
Upload pruned model finished
04-10 : v2.2
I attempted to improve the skin tones and lighting and shadows, which were previously awkward. I have noticed a difference in the image, but I have not had the opportunity to thoroughly test the changes due to my limited skills.
VAE : vae-ft-mse-840000-ema-pruned
Steps: 50, Sampler: DPM++ SDE Karras, CFG scale: 20, Size: 512x768, Model hash: a8907e32c0, Denoising strength: 0.35, Clip skip: 2, Hires upscale: 2, Hires steps: 20, Hires upscaler: R-ESRGAN 4x+, Dynamic thresholding enabled: True, Mimic scale: 10, Threshold percentile: 99.95, Mimic mode: Cosine Up, Mimic scale minimum: 0, CFG mode: Half Cosine Down, CFG scale minimum: 15, Eta: 0.2, DDetailer prompt: "", DDetailer neg prompt: "", DDetailer model a: "bbox\mmdet_anime-face_yolov3.pth [51e1af4a]", DDetailer conf a: 30, DDetailer dilation a: 4, DDetailer offset x a: 0, DDetailer offset y a: 0, DDetailer preprocess b: False, DDetailer bitwise: None, DDetailer model b: "None", DDetailer conf b: 30, DDetailer dilation b: 4, DDetailer offset x b: 0, DDetailer offset y b: 0, DDetailer mask blur: 4, DDetailer denoising: 0.35, DDetailer inpaint full: True, DDetailer inpaint padding: 32, DDetailer cfg: 11
04-01 : v2.0
As always, I didn't conduct many tests. However, I did notice a slight improvement in prompt recognition.
I'm not entirely certain if it's an actual upgrade from the v1 version or not.
Steps: 40
Sampler: DPM++ SDE Karras
CFG scale: 20
HIRES. FIX
R-ESRGAN 4X+
STEP : 20
DENOISING : 0.35
UPSCALE : 2
Dynamic Thresholding
MIMIC CFG SCALE : 10
PERCENTILE : 99.95 OR 99.9
MIMIC SCHEDULER : COSINE UP
MINIMUM VALUE OF MIMIC : 0
CFG SCHEDULER : HALF COSINE UP
MINIMUM VALUE OF CFG : 0
Ddetailer
bbox\mmdet_anime-face_yolov3.pth
Denoising strength(Inpaint) : 0.35
Mimic CFG Scale : 15 (custom ddetailer setting)
----------------------------------------------------------------------------------
03-17 : v1.0
I have updated the model with the extension "safetensors" upon the request of someone else, while keeping it the same as the existing model.
----------------------------------------------------------------------------------
This is a realistic model that I personally merged.
There is limited information due to a small number of tests conducted.
This is an example of a prompt used
ultra realistic details, sharp focus, detailed skin, (21 years old pretty korean girl:1.2), (kpop idol:1.2), (korean beauty:0.8), (korean mixed),
(large breasts),
Negative prompt: Drawings, abstract art, cartoons, surrealist painting, conceptual drawing, graphics, (low resolution:1.4), (blurry:1.3), (strabismus:1.1),
(worst quality:1.3), (low quality:1.3), (thick thighs:1.2), collage, (makeup:0.8), nsfw, bad proportions, earings, floral print, loli, big eyes, (watermark:1.2), letter, (abs:1.2),
Steps: 40, Sampler: DPM++ SDE Karras, CFG scale: 7.5, Seed: 3203463706, Size: 768x512, Model hash: 1d8d6e1204, Denoising strength: 0.35, Clip skip: 2, Eta: 0.2
HIRES. FIX
R-ESRGAN 4X+
STEP : 20
DENOISING : 0.35
UPSCALE : 2
Dynamic Thresholding
CFG : 20
MIMIC CFG SCALE : 5
PERCENTILE : 99.95 OR 99.9
MIMIC SCHEDULER : COSINE UP
MINIMUM VALUE OF MIMIC : 0
CFG SCHEDULER : HALF COSINE DOWN
MINIMUM VALUE OF CFG : 10