The GN version is still in testing and is not recommended
It is forbidden to sell the images generated by this model for profit, and it is forbidden to sell this model. Thank you
禁止将此模型生成出的图片进行出售盈利行为，禁止出售此模型。谢谢
Look here!!
typesmallv2
<lora:super-choko-Gnsu-rh-ve1:0.7>,typesmallv2, choko, 1girl, solo, bloomers, slit pupils, animal ears, cat ears, floral background, underwear, dress, tail, twintails, socks, frills, open mouth, flower, smile, shoes, cat tail, long hair, :d, looking at viewer, bow, bangs, black hair, sitting, blunt bangs, holding flower, blush, kneehighs, frilled dress, holding, hair bow, :3, long sleeves, full body, white bloomers, brown eyes, sailor collar, cat girl, pink dress, sailor dress, ribbon, knees together feet apart, frilled shirt collar, knees up, head tilt, hair ribbon, mary janes, sidelocks, footwear bow, hair ornament, animal ear fluff, fang, heart, outline, frilled sleeves, black footwear
请阅读模型的关于此版本
Please read about this version of the model
About this version
The model generalizes and transfers very well, and you can use it on almost any submodel (except SDXL).
If you want to restore the original painting style, you need a lot of prompt words to correct, the model's data set contains all nekopara images, the game illustrator artbook and fanclub cover, theoretically it can restore all images
Description
Wait for later version
FAQ
Comments (16)
Do you plan to update this model in the near future? The new version quite often gives whacky results and poorly listens to prompts
400,000 steps...
that certainly makes LoRA overcooked...
idk how many images there are in the dataset, but 400.000 steps i'd say would be nice for a checkpoint and not LoRA
It is still being updated, because I am currently sending pictures on pixiv, v4 version has poor nsfw support, so I trained this version, currently it is only used by me to generate nsfw images, of course, it is still updating iterations
You can use the v4 version first, at least it is normal
@LittleJelly I have tried dreambooth, but the picture is very strange, so I am obsessed with lora, and 400000 seems to be a common thing for my lora, almost all my lora have more than 100,000 training steps, and some su versions even exceed one million, but they still work normally. And there's no sign of overfitting at all
@LittleJelly And I'm only obsessed with cat girl, I don't seem to be interested in the others, and I have a lot of data sets that I'm ready to start training nekopara when I'm fully trained
it's not about obsession at all...
but using so many steps for one LoRA is an overkill...
rather make a Nekopara checkpoint/model with all charas included, it will be much better and we will be able to generate any Neko we want
@LittleJelly I tried dreambooth, but the performance was not as good as lora
@LittleJelly And for the overfitting problem, I used multi-model fusion to solve it
@LittleJelly I'll try dreambooth again today
Attention!
The v7 model is not recommended at this time, it is just a version that I am merging with the v4 version using the base model I trained, and currently Chocolate lora is updating iterations and preparing to solve the conceptual confusion (which I already have a way to solve), which is expected to be completed in one to two weeks. Can I have more backpictures and comments, let me know what are the disadvantages of this model
I have too many versions on hand, so the naming is confusing qwq
Gn is the concept
rh is for fusion
su is the new version
I have nearly 200 files in my chocolate lora alone
Would you consider uploading a video showing how you managed to generate something like the last image? I tried messing around with some of the metadata but could not recreate it at all
You can put your generated image on the post below and I will solve it
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.