MeinaMix objective is to be able to do good art with little prompting.
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
MeinaMix is officially hosted for online generation in:
- SeaArt
- Mage.space ( with animate feature )
MeinaMix and the other of Meinas will ALWAYS be FREE.
Cover image lora made by: FallenIncursio | Civitai
Recommendations of use:
--------------------------------------------------------------------------------
Enable Quantization in K samplers.
Hires.fix is needed for prompts where the character is far away in order to make decent images, it drastically improve the quality of face and eyes!
---------------------------------------------
Recommended parameters:
Sampler: DPM++ SDE Karras: 20 to 30 steps.
Sampler: DPM++ 2M Karras: 20 to 60 steps.
Sampler: Euler a: 40 to 60 steps.
CFG Scale: 4 to 9.
Resolutions: 512x768, 512x1024 for Portrait!
Resolutions: 768x512, 1024x512, 1536x512 for Landscape!
Hires.fix: R-ESRGAN 4x+Anime6b, with 10 steps at 0.3 up to 0.6 denoising.
Clip Skip: 2.
Negatives: ' (worst quality, low quality), (zombie, interlocked fingers) '
Negatives if you can't use Hires.fix:
'(worst quality:1.6, low quality:1.6), (zombie, sketch, interlocked fingers, comic)'
--------------------------------------------------------------------------------
In the merged models list: MeinaMix V1~11, MeinaPastel V3~6, MeinaHentai V2~5, Night Sky YOZORA Style Model, PastelMix, Facebomb, MeinaAlterV3 i do not have the exact recipe because i did multiple mixings using block weighted merges with multiple settings and kept the better version of each merge.
Description
This is the same as the original V10 but in different format.
FAQ
Comments (53)
Is there a significant difference between this and your Hentai model? Couldn't I use either models to create sfw/nsfw images?
MeinaHentai does work for generating sfw and nsfw but in my testing MeinaMix refused with all of its will to generate images that are expected from hentai without lora, while MeinaHentai works well for that even without Loras
Like your models, why dont make a inpainting model?
when i tested inpainting model vs normal, the difference was so little that idk if its worth the space it takes x.x
@Meina even less worth the time now that there's a controlNet inpaint model that improves inpaint significantly for all checkpoints
@warMonkey What is that model?
@jdfjhsakhfsahjr It's the inpaint model that comes with Controlnet. Not sure what your question exactly was but controlnet is an extension that you can install on A1111, go to extensions tab then "available" search for it and install. if you can't find it there for some reason, search for it on github and install manually from extensions tab.
Is there an Inpainting version?
Do you have a model that does not change the background so much if, for example, there is a white wall behind?
hmmm, what prompt have you been trying? because if you add something like 'wallpaper' then it add details into the background
This is my favorite checkpoint, thank you so much for making it and sharing it
I tried many anime models and by far this (along with some of your other meina models I tried) is the best! Even with all these downloads recently I think this is still underrated. Thanks for making it!
I know you said in the other comment that you didn't see much difference between inpainting and non-inpainting version of the model but I do think it's still worth it to create one as they are much better at out-painting than normal ones. I wish you can make one as this is the best anime model so it would be complete with an inpainting model :) Thanks again.
Thank you for your words!! q(≧▽≦q)
About the inpating, i don't want to do it rn and bait update so i'll do it for V11! however i can do for V10 and upload it to huggingface only
@Meina Thanks a lot! We can wait for V11 if you don't want to do it now, no rush :)
@Meina ye
1111
1111
Any chance for unpruned fp32 with unbaked vae?
In some cases result differ enough to warrant trying.
getting some blurry results? any ideas how to fix?
do you mind joining discord.gg/meinaverse and creating a topic in the Help channel? so its easier for me to look into your problem and fix it :3
I'm pretty sure a VAE could help
@atypicaldayin2023921 my models all have VAE already tho
@Meina They might not have enough memory..? Sometimes I get blurrier results the more memory the model takes up. I only have 7gb of usable memory :(
i got a VAE from another comment with a similar problem and it fixed the problem
I dont know what promt should I used
any prompt you want, you just have to type what you want SD to generate :3
If you need ideas, head on over to gelbooru. You can also research particular types of dresses, accessories, hairstyles, etc., as you encounter them from watching movies, anime, TV series, and so forth.
Supposedly this model has a built-in VAE right? Well, I'm getting very dark images and I don't know why. My generation data is:
beautiful scenery, fantasy theme, magic castle, rainbow, particles of light, sunbeams, rainbow, field of flowers
Negative prompt: (worst quality, low quality:1.4), (zombie, sketch, interlocked fingers, comic)Steps: 60, Sampler: Euler a, CFG scale: 11, Seed: 1436931507, Size: 1024x512, Model hash: 77b7dc4ef0, Model: meinamix_meinaV10, Denoising strength: 0.3, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+ Anime6B, Version: v1.4.0
The results: https://imgur.com/FW0qkCd, as you can see it's dark.
It does have baked VAE as for the cold colors, you can try using something like this lora in the prompt but use negative weight for it https://civitai.com/models/13941/epinoiseoffset like <lora:loraname:-0.5> and that should make the colors be more warm
@Meina Nope, didn't work. :(
@tsukana ＞︿＜ , do you mind trying with meinapastel v6? it have warmer colors than meinamix v10 by default, i added too much noise in V7 of meinamix and have been slowly making it more warm with every update
@Meina That's better, thanks.
Hello, just a question. I'm always using LORA for making characters... lets see an example, Momo from To LOVE-Ru. We know that To LOVE-ru has that cute original manga trait, but when I run LORA it becomes more my LORA trait than the original trait, how can I avoid that and try to be more like the original design?
Helloo!! the only way is playing around with the lora weights or using another lora D:
In my discord theres a little tab for some lora creators that make lora tested in the meina models and theres one of the character you want, maybe you can give it a try? https://civitai.com/models/57991/momo-belia-deviluke-to-love-ru
ヾ(≧▽≦*)o
Or you can try to change the weights of individual blocks of that LoRA using this extension: https://github.com/hako-mikan/sd-webui-lora-block-weight
Great links, thank you)
pruned 2gb version plz
You can use model converter extension to prune it yourself. Settings: FP16; safetensors; pruning methods:disabled; extra options -> others:delete. Then just press "Run"
You can also read my article on this topic
@Miclasher not all ppl use automatic111
専門的な世界に入りたいです
"MeinaMix objective is to be able to do good art with little prompting."
I suppose that is why I both love it and hate it, hahaha
Sampler: DPM++ 2M Karras: 20 to 60 steps.
What if i do more? Like a 100? Will it make it glitchy?
best model (along with goofball) , meina's toes are the best
95% of the images ive produced i had to switch to meina to get the toes right
What the fuck? It's the same thing.
I know, my community told me about that, that other model is a merge of mine but with 0.001 in the OUT11 just so it would be another hash... https://imgur.com/2NVOe4l
You can check it yourself if you have the supermerger extension, theres an analysis tab there that compare the similarity between models!