5 is good

@PromptSharingSamaritan The problem must be that one, I've read that you were using a VAE and I downloaded it but honestly, I've never used one before. I put it in the embeddings folder thinking that was what I had to do, but apparently not. How can I use the VAE?

@PromptSharingSamaritan Thank you very much! I really appreciate it, I'll try it right now.

hi-res fix

I've found that this model benefits from hi-res fix

Use Hires fix ... it literally says so in the description.

look on the diffusers hub

@PromptSharingSamaritan where can i find the vae model

still crazy good though

@PromptSharingSamaritan thank you

现在解决了我的问题，在网上找方法调试了VAE模型，就修复了

@nikola_lin 你换的是哪个WAE呀？我也有同样的问题

@PromptSharingSam6319 Hello, I used your model and prompt words and found that the overall picture is a bit dark after use, and the face of the angel is a bit damaged. Is it convenient to ask what VEA model you use?

@PromptSharingSam6319 like those ones https://huggingface.co/models?library=diffusers

diffusers models are made to been trainable. But not all of them has quality to train faces.

I use TheLastBen Dreambooth Colab. If the model doesn't have diffusers it gives error

@PromptSharingSam6319 I have no idea hahah, never created a model from scratch

@PromptSharingSamaritan Thank you for replying! Also, did you simply merge multiple models by using checkpoint merger or did you use the generated data to finetune the new model?

@PromptSharingSam6319 Got it, thanks.

Hey! If you're using webui then edit the bat file called "webui-user" and add the following to the command arguments line: --reinstall-xformers --xformers --autolaunch

Then after your first launch you can remove "--reinstall-xformers" as that party is only needed to get xformers to donwload. xformers is a free performance enhancer for stable diffusion so you should be able to generate higher quality pics. If thats still not enough memory saving for you, you can add "--lowvram" to the argument line as well though this makes it take longer.

I recommend ADetailer!

It will correct the problem nicely.

I recommend ADetailer!

It will correct the problem nicely.

Thank you! I'll give it a try. Which upscaler, and how many hi-res steps do you use?

@artydelly247 try sd up

@PromptSharingSamaritan there is 4-5 of them there.. which shall I get? Also I noticed they are "vae-name.bat" type files?? shouldn't they be *.ckpt files or something similar? a *.bat file is a executable file, which is kinda suspicious.

@PromptSharingSamaritan YOZORA.vae.pt is not showing up in the vae options after I downloaded it to the vae folder

@PromptSharingSamaritan I downloaded three VAE's from civitai and even after selecting them in the VAE option I am still getting pale colors. the colors are bright during the generation process but once the generation finishes the colors turn pale. Please help me I really want to use this amazing model

https://huggingface.co/Fallere/MMK-E/resolve/main/YOZORA.vae.pt?download=true

@PromptSharingSamaritan can you please tell me how to correctly use VAE. I downloaded one and put it in the vae folder and after that I selected that vae from the option in the UI but it's not working

@nitinharshitalucky212 maybe you using the "built-in" vae in the model?? ...and make sure you are using a SD vae for SD and SDXL vae for SDXL.. but yeah, just dropping it in the vae folder and loading it in the workflow is all you should need to do.

@nitinharshitalucky212 thats not a vae issue... Sounds like a denoise issue

@jerimy777 How can I resolve that?

@nitinharshitalucky212 well just change the denoise level.. try .7- 1.0... message me the list of your settings and models.. Ill try it on my end and see what I can come up with for you. It's hard to say without any info.. read the command line when que and new image. Sometimes it will still show an image but ran into errors while creating. 

@nitinharshitalucky212 you have to restart StableDiffusion before they will populate the list of. Make sure to unzip the file first.. But Vae typically have little effect on the output anyway. I wouldn't worry with it too much.

@PromptSharingSamaritan u should link it here

UPDATE: I made an inpainting model using this thread I found on Reddit: https://www.reddit.com/r/StableDiffusion/comments/zyi24j/how_to_turn_any_model_into_an_inpainting_model/
I highly recommend anyone to try it!

Haha download and try it out I would say

No, but I remember kl-f8-anime2.ckpt and YOZORA looked identical at the pixel level.

@Zenyth damn, can't import to invoke, Error: Model install error

Cannot determine base type