MeinaHentai Objective is to create NSFW illustrations without much need of Loras!
I have a discord where you can share images, discuss prompt and ask for help.
https://discord.gg/meinaverse (✿◡‿◡)
我有个可以让你分享图片和参与讨论与询问问题的discord群。
I also have a ko-fi and Patreon page where you can support me or buy me a coffee <3 , it will be very much appreciated:
https://ko-fi.com/meina and https://www.patreon.com/MeinaMix
MeinaHentai is officially hosted for online generation in:
- SeaArt
- Mage.space ( with animate feature )
----------------------------------------------------------------------------------
-- Recommendations of use:
If you want the art to be more realistic, add: ' realistic ' or ' photo ' in the prompt.
---------------------------------------------
Recommended parameters:
Sampler: Euler a: 20~40 steps.
Sampler: DPM++ SDE Karras: 20~30 steps.
CFG Scale: 7.
Resolutions: 512x768, 512x1024 for Portrait!
Resolutions: 768x512, 1024x512, 1536x512 for Landscape!
Hires.fix: R-ESRGAN 4x+Anime6b, with 10 to 15 steps at 0.2 up to 0.4 denoising.
Clip Skip: 2.
Negatives: ' (worst quality, low quality:1.4), monochrome, zombie, (interlocked fingers:1.2), '
--------------------------------------------------------------------------------
Description
Inpainting model for Hentai V4
DO NOT USE IT FOR TXT2IMG :3
FAQ
Comments (36)
Hey I just wanted to thank you for publishing this model, it's one of my favorites! I'm using it right now as a base for training my LoRAs.
how do you train custom lora with a checkpoint model ?
@abinpramdya2922 Hi! I recommend you looking into youtube videos about lora training with Kohya_ss, also join the Civitai discord server, there's a channel dedicated to help with chekpoint and lora training.
why did u not bake the vae in it? ur results with baked vae are misleading
what do you mean? i only upload the models after i bake the VAE in them
i even opened it in toolkit to verify if something was wrong but nothing is wrong, its there https://imgur.com/a/zZcx44z
Maybe you have VAE overload set in your webui?
Can generate images of semi-furry/kemono characters?
Can you please upload a XY plot comparison of V2.1 V3 and V4?
Why the heckkkk does your hentai models look better than most checkpoints lol
Where is V5??
i uploaded V5 only for patreons to download, they get early access to my models as a way of me to thank them for supporting my work
Yeah I have no clue how to get my images to look that good, I'm literally using the same prompts, seed, and everything, but mine come out looking bad, I've tried everything. I've read some of the comments and it seems you're using a totally different model then what's advertised. I really don't think that should be allowed.
use hires with upscale x2
Why... would I do that if I have the generation data in all of the images? And the parameters I use in the model description to get the same results as i do?
But all of the images I generate is using the automatic1111 webui running locally on a Nvidia GPU. And since the webui influence the generation directly some of the images will now by impossible to get 1:1 generation unless if you roll back the webui to the version of months ago which is when the model was released.
And the code sometimes is different between the showcase because I do the testing in a raw model without baked VAE to see which VAE I will bake in so new users don't have to worry about doing testing themselves.
But if you want help with your generations, you can come to discord.gg/meinaverse and my community or I will try to help you! 💜
A lot of the prompts are "Eula a", make sure you don't have accidentally changed your ENSD (Eta noise seed delta) changed. Most models use 0, but some use like 31337. Its not show in the civitai prompts, but often can be seen in the images meta data, if applicable.
Also make sure which random number generation is set. CPU and GPU create very different images. Newer Automatic1111 also has NV for RNG, which very closely reproduces the RNGs of nvidia GPUs (the GPU setting).
So if you had it on CPU, try NV or GPU instead etc.
You gotta tweak with some of the stuff yk
There are some settings haven't been displaied by the civitai, you may try download the png image, use the "PNG Info" to read full settings from it and then use the "Send to txt2img" button to apply those settings and finally hit the "Generate" button.
I can reproduce the exact same image with those steps.
And thanks @Meina again for providing those excellent models to us :D
Are these based on Anything3 / AbyssOrange in any way?
Your stuff is super amazing but I've noticed older LORA's trained heavily on those old checkpoints sort of go wonky on newer stuff.
TL;DR -
I'm looking to clean out my model directory of older checkpoints and replace them with (IMO better) stuff like this.. maybe I should just test some older LORA's on this and report back. :thinking:
they're not based in aom or anything3, thats why some loras are kinda weird D:
Any chance V5 will be posted here or is it forever exclusive to Tensor.Art and not downloadable.
Hellooo, i'll see how it goes but for now the only way to download V5 is through my patreon, since they always get early access to my stuff and the tensor is also a test that i'm doing, but eventually i'll publish it for free since i've seen saying my models are always gonna be free ( if you don't mind waiting a bit )
@Meina Its been way over a month and still no V5
Maybe a please would help.
How much longer lol, some of my tensor art is not displayed here properly since this place only has v4
DUDES!!! “ for now the only way to download V5 is through my patreon”
NOTE: v5 is on Patreon
───[ VAE ]───
MSE-840000: Seems to improve slightly.
♦ Correct the shape of the eyes and minor details a little.
- It takes away a little definition (almost imperceptible)
Waifu Diffusion: It is one of the few models that seems to work with this VAE.
This vae distorts the lines quite a bit and intensifies the colors.
Compared to other models. This model maintains the shape of the eyes quite well when applying the Waifu Diffusion vae.
(This vae is only occasionally recommended)
Summary.
MSE: Slight improvement
Waifu Diffusion: Works very well (considering the difficulty of this vae)
Note 1: It doesn't really need vae, and it may work better for me because I try to put minimal prompts (+ EasyNegative v1).
Note 2: Translated from Spanish (Spain) to English. Therefore, it may contain small translation errors.
What is the base model that this Checkpoint is trained on?
Still got a blcoked by provider message on all images
Are you gonna make an XL version?
6
Does cuppy know you do this? LMAO
nice!
There are many errors at hand
It’s confusing
Pretty cool!