Disclaimer
v3 showcase images are done with full Dreambooth, but model provided is an extract. Some inconsistencies with them are expected, but not major.
XL version - https://civarchive.com/models/276075
Kofi - https://ko-fi.com/anzhc
And i guess im opening my request form - https://forms.gle/8bhruFiecpy5szNf6
v3.1 watermark fix: This version improvement is cleaned up dataset. Some bad images were removed, 209 watermark apparitions were erased. Training was also done more finely.
v3: Training done as a Dreambooth, and then extracted to rank 128 locon.
v2.5: more images that are meant for aesthetic purposes(tagged accordingly). Training changed.
v2 update: 682 image training, but only 10 epochs, and on other base model. Should work quite good.
Full version update: Training on 682 images, vs 200 in Lite version. More dreamy, more abstract, better eyes, for whatever reason. But composition is less, let's say, tangible, in a way. But that's the point of this model i guess :D
General instruction: try some bullshit, random words, something dramatic, use wildcards. It will react differently. By default style is not very interesting(unless it is), it needs some prompting to get going.
--FOOTER--
Model provided strictly for non-commercial use.
Reprint(repost, reupload to other sites) is prohibited.
Description
Trained with 682 images vs 200 in Lite.
More dreamy, eyes look much better. But less tangible than Lite.
FAQ
Comments (11)
What is the Epoch that you publish, so high weight and so many epochs will not overfit
What GPTS stand for?
up,GPTS4云朵艺术和RPG MODEL 非常搭配，可以去试试看。
优秀到大家集体隐藏图片不发表！ps：up加油，作品无敌，GPT整个系列都形式感极强，你迷恋的粉橙色系统使很多经典底模再次闪闪发光！
can't get lycoris to work
It's very beautiful
(づ ◕‿◕ )づ
This is an abstract question, but I want to describe various things around the characters, like small objects, backgrounds, etc., but it doesn't work.
How can I create a picture that is denser as a result?
It looks difficult to get rid of the text in the images, negative words also do not seem to have an effect. Any suggestions?
Apparently, it takes an unusually long time to process under certain conditions.
It took 11 minutes for hires.fix to enlarge 512*768 by 1.5 times (768*1152), and then 15 minutes for Loopback Scaler*4.
It seems that detail-increasing LORA is not necessarily compatible.
(Abstract Dreamwave to do detail->another LORA tries to increase detail further->Nightmare)
I tested this out for a bounty, and i have to say that i am genuinely happy with the results you can get from it.
Using MeinaMix, short prompts and a strenght of :0.7-1, you can generate pretty good images, that doesn't look too generic.
In this case, it was used in conjunction with LORA, which is a hair-split.
It was excellent because there was no speed problem, and the LORA correction was applied to "the area around the hair" and behind it.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.