To support my work, you could buy me a coffee
https://www.buymeacoffee.com/JasonAICreator
Prompt Parameter Suggestion:
Style tag:
real life,instagram,
Main Prompt: 1girl,
Portrait:
upper body,close up,portrait,
light:
cinematic lighting,high quality shadow, <lora:epiNoiseoffset_v2:weight>
Neg Prompt: (worst quality, low quality:1.4),(raw photo),nsfw,SimpleNegative
CFG: 7
Step: 30
Sampler: DPM++ SDE Karras
Clip Skip: 2
recommanded to use epiNoiseoffset_v2 to adjust darkness
Not recommand to use 'Restore faces'
CosplayMix4.2 [20230623 v4.2]
Merge 0.5 with Kawaii Realistic Asian Mix to adjust detail of generation
CosplayMix4.1 [20230618 v4.1]
Merge 0.1 fantasticmixReal_v30 to adjust detail of generation
CosplayMix4.0 [20230617 v4.0]
✍️Trained with 10k images using 4days with 3090
🫶Merge with realistic, anime models on different weights
🕹️Better tags control
Newly Merge Models:
freshPhotoRealism_v15
deliberate_v2
cosplayMix3
The aims of this model
To create a realistic model which could perform better cosplay with anime character on my aesthetics.
Commercial use
This model doesn't allow any commercial use by selling the model, hosting the model on platform.
You could use this model to generate images and take your own responsibility.
CosplayMix3.0 (abandoned)
Completely new merge with different models.
To use version 3 for photorealistic
Don't recommand to add a lot of quality tags, such as
❌8k, intricate details,ultra-detailed,absurdres,real picture, (photorealistic, hyper-realistic:1.2)
If you just want some "quality" images, you might still use it. (gallary images still use quality tags just because I am lazy to adjust and for the comparison with v2)
Prompt suggestions:
👍Positive prompt prefix:
natural light,
👎Negative prompt prefix:
(raw photo), EasyNegative, badhandv5, (worst quality, low quality:1.4)
Neg promp embeddings:
https://civarchive.com/models/30452/bad-hands
https://civarchive.com/models/7808/easynegative
[20230606 v3.0]
Aims add more photorealistic and decrease the AI feels
Might be adjust the model later like 3.1, 3.2, etc
New merge based on basil mix.
Reduce the % of chilloutmix.
Reduce the % of nsfw content which mainly came from grapefruit-hentai-model
But added a little bit cosplaymix2.0 back to the new merge
which keep the taste of cosplaymix.
(Cosplaymix mainly based on chilloutmix)
Newly added RunDiffusion FX 2.5D (not photorealistic
https://civarchive.com/models/82981?modelVersionId=88167
Recommaned models:
RunDiffusion FX Photorealistic
https://civarchive.com/models/82972?modelVersionId=88158
RunDiffusion FX 2.5D
https://civarchive.com/models/82981?modelVersionId=88167
CosplayMix2.0
I generally satisfy with this version.😆😆😆
If you feels it is a good mix, please comment✍️ and leave an images 🖼️. Both are appreciated.
If you like this mix, please give it a heart❤️.
This mix aims to keep photorealistic and capable with most anime lora.
[20230418 v2]
version 2
🚀Improvement Highlight effect.
🚀Improve Realistic skin texture
Merge List:
https://civarchive.com/models/6424/chilloutmix
https://civarchive.com/models/24383/grapefruit-hentai-model
https://civarchive.com/models/20981/3moon-nireal
https://civarchive.com/models/22402/fantasticmixreal
v1.1 added
https://civarchive.com/models/22922/lyriel
https://civarchive.com/models/9291/sunshinemixsunlightmix
v2 added
https://civarchive.com/models/15003/cyberrealistic
https://civarchive.com/models/17754/dreamful
Suggested steps: 20~35
Suggested CFG: 4~9
Some example images using extention: Dynamic Thresholding (CFG Scale Fix)
Description
FAQ
Comments (20)
What VAE is the best? Like i copy&paste prompts but the results still do not look realistic.
840000
If you copy and paste prompt, you might missed some neg embedding
很好的模型,请问怎样才能联系到您
商务微信:fkdeai,学习交流qq群: 866612947
can you reupload the version 3.0 please? I really like the cartoony looks of that model!
You might use 2.0 version instead. 3.0 version is pretty bad. Usually generate bad hands. The different between 2.0 and 3.0 is chilloutmix which is reduced a huge percentage in 3.0. 3.0 use similar merge with 2.0 but replaced chilloutmix with basilmix.
@xiaozhijason Thank a lot! I will try it out later today!
其待之後更新~
are you using hires,fix to make your result more HQ? I'm still dont know about how to make the result become HQ
Yes, most images are used hires fix.
@xiaozhijason and what is the amount of denoising scale so that the result looks perfect?
@oppanakal626 usually 0.3~0.4 for t2i. It would remains the original image and upscale.
Great work!! Have you ever considered to make a SDXL version?
I am making a sdxl model named openxl. In the further development, it would considerate cosplay costume training. But not the first priority.
@xiaozhijason Appreciate for your reply. I've also noticed "openxl". Hope everything goes smoothly!
Are there any disadvantages of v4.0? Why is there no download for v4.0? --- Google Translate
v4.0 is removed from the page.
總覺得很不好出黑絲過膝襪誒,基本會衹剩下個腿環,或者沒有黑絲,要麼就是黑絲材質不對。
looks cute







