v6.0 and v6.0 Sharp released simultaneously
About this version
[[[Adjustment environment]]]
Sampling method [ DPM++ 2M Karras ] [ Euler a ] [ DDIM ] ets
Sampling steps 14~30
CFG Scale 4.5 ~ 8
@negative prompt
EasyNegative, (worst quality:1.5), extra fingers, fewer fingers, bad-hands-5, bad-artist, bad-artist-anime, (photorealistic:0.28, realistic:0.4, 3d:0.25), render, ((censored)), rose, ((animal ears)),Unspeakable-Horrors-Composition-4v, verybadimagenegative_v1.3, negative_hand-neg, bad_prompt
or
worst quality,low quality,normal quality, bad-picture-chill-75v ,EasyNegativeV2
or
EasyNegativeV2
/ / / / / / / / / / / / / / / / / / / / / / / / / /
=Hires. fix=
Denoising strength 0.5
Hires sampling method DDIM or DPM++ 2M Karras or Euler a~
Upscale by 1.5 or 2 ~
Hires steps 20
/ / / / / / / / / / / / / / / / / / / / / / / / / / /
Extensions that helped
[ FreeU ] [ Ranbooru ] [ LoRA Block Weight ] [ SuperMerger ]
FreeU Integrated,SelfAttentionGuidance Integrated,Kohya HRFix Integrated,LatentModifier Integrated on SDforge
in my environment VAE
vaeFtMse840000Ema_v100.pt
color101VAE_v1.pt
using.
The reference image was generated using this.
Please use your favorite VAE.
First of all, I have no knowledge of training. My theme is exaggerated expression, or expanded color.
先に言っておきますと、私にはトレーニングを行う知識がありません。誇張した表現、つまり拡張した色彩が私のテーマです。
If possible, please upload at least one photo. I want to see it too.
Regards XD
Description
v2.0は、もっともダウンロードが多かったコントラストバリエーションGEXtenB60-75を元に、loraを厳選し1つマージしました。
さらにweakened unityのプロトタイプと丁寧に編み込みました。
結果使いやすい形に収めることができました。やはり癖はあるものの予想できうるレベルでの操作性を持ち、なおかつ芳醇な質感と、適度な遊び心でいい意味で期待を裏切られたモデルが仕上がりました。
全バージョンは複雑な操作性が必要でしたことを反省いたしました。よってloraの選別には慎重に慎重を期しました。その点においては、私は自信をもってこのモデルを広くなじみを持っていただけるものと考えております。
勿論アップデートでありますので、前モデルの奥深さかも損なわれないように気を付けましたが、それを主張できるテストは今回は行っていません。ただ祈るばかりです。
v2.0 is based on the contrast variation GEXtenB60-75, which was the most downloaded, and has been merged with one selected lora.
Furthermore, it was carefully woven together with the weaked unity prototype.
As a result, we were able to put it into an easy-to-use format. Although it has its quirks, it has a predictable level of operability, has a rich texture, and has a moderate sense of playfulness, resulting in a model that exceeds expectations in a good way.
We regret that all versions required complicated operability. Therefore, we were very careful when selecting lora. In that respect, I am confident that this model will become widely popular.
Of course, since this is an update, we were careful not to lose the depth of the previous model, but we have not conducted any tests this time to make that claim. I just pray.
FAQ
Comments (7)
11.1G.等修修版了，当年Zovya Photoreal ultra的初次发布也是11G惊为天人。然后给其它玩家修到1.8G，不过现在都是4，5G的正常模式了。这个还有修修空间，以后再来。
"Complexly tuned with several models and many loras."
...and you have just only one image to show?
I'm happy you are so proud of your model, but I'd like to know more information about to use it.
又来评价了。七合一，有点东西
GEXtenシリーズはコントラスト調整のテンサー割り込み違いのバリエーションになます。
The GEXten series is a variation with different tensor interrupts for contrast adjustment.
(referencing to v2.0)
Sure the demo gallery looks fantastic. But this checkpoint didn't work well for me.
The problem is when you try to define an age in the prompt. At this point, it will do wrong the body proportions, mixing adult parts with young proportions.
I think the checkpoint should be better trained to handle this...
推奨するVAEはありますか？
download VAE?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.