add no people in positive and make it in double parentaseases

negative prompt only works for image generation

if you want to get rid of actual people you need a different approach

It's a VAE issue. Use the WD VAE like described in the description. (The updated SD 1.5 VAE, and other anime fine-tuned VAEs may also give good results.)

It's because you're not using VAE. You can download it from the dropdown on the right side of the download button.

additionally try borrowing someone's image prompts

Do you have the VAE in the same folder as the model? It should be in the VAE folder, not in the stable-diffusion folder. You can select what VAE to use in settings/stable-diffusion (assuming you're using Automatic1111)

@theVoidWatches i figured it out!! just forgot to update lol. thank you

A little late, but you can find people that will help at the Unstable Diffusion discord: https://discord.gg/unstablediffusion

Is it just me or everytime I conform discord invitation I got beep boop screen then instantly got banned. I've tired with so many accounts with instaban result.

Pro tip: avoid discord.

Yes you can, you just need a lora for the specific character.

You can try setting the VAE to always be the pastel one via the settings if you are using Automatic1111, since I've had issues with the automatic pickup

rename the VAE to the same name as the checkpoint (except for the extensions). Automatic detection usually uses this simple rule.

 wait so we put it all on the stablediffusion file right?

or the VAE should in the VAE folder (sorry i'm new to this) @ZeroEight 

@s01106 you put the vae in the same folder as where you point the checkpoints. If you want it to be picked up automatically, you need to rename it to the name of the checkpoint you are using. Otherwise, you need to to the settings tab and search for VAE, there is a dropdown where you specify the VAE to be used (default is automatic).

I found a solution to your problem. You need to enable in the settings and in stable diffusion not automatic, but the one you want. Change the name of VAE and Chackpoint in advance, so that they are the same.

Then turn on chackpoint.

Belisimo!

I found a solution to your problem. You need to enable in the settings and in stable diffusion not automatic, but the one you want. Change the name of VAE and Chackpoint in advance, so that they are the same.

Then turn on chackpoint.

Belisimo!

If you want to draw a character from the "touhou" genre, you need to find the character's lore first in order to draw them accurately.

Have you found any solution to this? I have the same problem

@Mercus I didnt, i downloaded another pastel model from hugging face which includes in itself vae

Called Pastel Better Vae

I found a solution to your problem. You need to enable in the settings and in stable diffusion not automatic, but the one you want. Change the name of VAE and Chackpoint in advance, so that they are the same.

Then turn on chackpoint.

Belisimo!

"Fantasy.ai is the official and exclusive hosted AI generation platform that holds a commercial use license for Pastel-Mix, you can use their service at https://fantasy.ai

Please report any unauthorized commercial use.

Huggingface repository goes here.
Interested in supporting me? Buy me a coffee."

With regards to coffee, Fantasy.ai now is the official and exclusive platform that should buy Pastel Mix coffee.

Right click the download button and paste it the custom url section, join here for more :)

https://discord.gg/touhouai

webui-modles-VAE

well hate fantasy.ai as much as you want but wtf the model is not crappy at all

U download the VAE and put it inside the Models/VAE similar that with the stable-diffusion models, then go to setting click "show all pages" ctrl+f look for "SD VAE" then select the VAE make sure to click refresh next to it if you dont see it, then MAKE SURE to go back all the way up and click "Apply settings" and you good to go, the VAE should be working now and apply better colors

There's two forms of generation involved in getting the sort of stuff you're seeing (and techniques inbetween). Firstly, try simple prompts describing what you want to see in the scene using singular words separated by commas (this is typical of most anime models, most don't want to be told a story in sentences of what you want to see).

Next is resolution. Try either 768px in either length or width (or both). The higher the resolution the better (sometimes, many times it can fail since Stable Diffusion 1.5 models were mostly trained with 512x512 images), but try to keep it in multiple of 64 if you can. If you don't want to natively generate something in higher resolution, go with the HiRes multiplier while starting with the base resolution 512x512. Go for a multiplier of 1.5x, or 2x (the higher VRAM you have the higher you'll be able to push this). You need to also hunt down a download of the upscaler that'll be used (many use R-ESRGAN 4x + Anime6B). There are many other though.

Next, make sure you're using DMP++2M, or DMP++2M Karras as the sampler, this seems to be a default good choice for 2D works, but try out others to see the stylistic change. Be aware, not all generate at the same speed, or do good work at lower sample steps. If it's producing lots of artifacting, try more sample steps. The prior two I mentioned begin to show great results after 20, and really solid results by 40 steps.

Try with and without VAE loaded, usually some models have a VAR baked in since most users get washed out images when using a model that hasnt had a VAE baked into it. The problem with ones that do have baked into it, is while they do provide solid colors and line work, loading another VAE of your choice is most times going to ruin your output. So try all models with or without VAE (ones that dont say if they come with or without one baked in by the model maker, will usually be color desatured if they didn't have a VAE in it which will allow you to load yours).

There is the next step I wont mention since all this is enough for you to take to Google/YT and learn more. But that's something called inpainting. That's the final stage of seriously good looking generations.

Really funny as a system named fantasy sys~

yes you can, people have been doing these for all models since SD was released, no one will know what model you're using

Hi! Here's a guide for you!

First of all, you need to have stable diffusion downloaded! I can't really help you from here with details, I am a noob too so I know it's difficult to follow but there are plenty of guides. Here are some things you should do:

1. Download Stable diffusion (here's a guide: https://rentry.org/voldy)
2. Once you've confirmed it runs, locate the folders "embeddings", "models" (inside of which you will find) "Stable Diffusion" and "Lora" (should both be inside models).

Now, in civitai everything is marked. If something is marked as "Textual Inversion" it goes into the embeddings folder. If it is marked as checkpoint trained or checkpoint merge, it goes into the model folder and if it is marked as Lora then it goes into the Lora folder. Basically, models are what allow you to shape the images, embeddings are some things that help with the generation by setting perimeters. Some really important ones like easynegative and deepnegative help to generate very good images, i use both with nearly everything, especially anime. Lastly, Loras can help you design a certain thing, like a specific character by using the triggers provided on the civitai page.

3. Now, with this new info, place the files in their respective folders. (I also recommend downloading Automatic1111 ui, as it is very helpful)

4. Search for prompts in the models you like, find some good ones, and generate!
Please do reply to me for more info as I know it is difficult to do, being a newbie myself. Hope I could help!

Oh yeah also forgot to mention VAEs! In the models folder you'll also find a VAE folder. Most models when you download them, have a VAE they come with. You need to download that and add it into the VAE folder. You can enable them through settings in the Automatic1111 ui. They're mostly used for finetuning but they're really important for most models. If a model doesn't contain a VAE its usually baked in, meaning its already implemented in the model without you having to download or run it!

the full version seems like already has a VAE (I just simply compared the size on Huggingface)

I can't find pastel-waifu-diffusion.vae. Can you help me with this please?

@Noire90 Just replied to someone else, here you go samle/sd-webui-models/pastel-waifu-diffusion

@Dreamer333 Thank you very much🙏 

bro's changing identity, might find him again if he makes something like pastelmixv4. lol

@EBIX No problem... too bad that some of the diffusions will be hard to find... hugging is nice but once it vanishes... we will wait and see :)

@malikxseto well, sdxl will be here after 2 days aswell, we made many friends and left many reaching here

Here's an archive of his models

https://huggingface.co/LMFResearchSociety/andite-finetunes-backup

maybe everyone point him "Anything fake creator" then he close all his work. but I think Anything v4 is best version btw.

@infrezz721 He deleted his HF for another reason which I will keep private to a certain point.
Main reasons for deleting is: Burnout, Thought no one used his models anymore, Wanted to get rid of his AI stuff.

@infrezz721 (anything v4 was a troll model, it was just a merge dont tell anyone)

@AshtakaOOf Good luck and best wishes to Andite.. Burnout sucks... I can relate... it leads you to a point that you want to give up everything.....

@EBIX It was a merge between one of his own model called Cocoa, available in the repo I sent above and AbyssOrangeMix2.

He trolled anything is one thing. At some point, he betrayed many AI model creators like me for money. Another story I don't want to go into detail. Consequence was, his models got removed on many image generator sides and he got outright banned on many discords.

@Ikena Are you talking about fantasy ai because if so he cut bridges with them a while ago

@AshtakaOOf Don’t want to go into details. Good for him

I can't find pastel-waifu-diffusion.vae. Can you help me with this please?

@Noire90 https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt

@AshtakaOOf Thank you AshtakaOOf. Are you sure this is the same?

@Noire90 Yes andite never trained a new VAE just for the model, and there's a reason why it's called pastel-waifu-diffusion.

@Ikena fillipino troll moment , trolled all ai creators, merged loras in model, became a base model for literally all the anime models in 1.5, made contract, became famous,got banned, disappeared , truely a legend, yet i precisely know where he is

@EBIX Must be fun to be a internet stalker

@AshtakaOOf not exactly stalking but he is in the official stable diffusion server just rarely appearing once in a while

yeah why did he private that?

@DevRev Basically he put himself into drama in an AI discord server and has chosen to move on from AI.

@AshtakaOOf damn, what kind of drama? I thought he was pretty chill guy, I lost one of his mix model and couldn't find it anywhere.

@DevRev You can get access to a repo with most of his mixes and finetunes above and more context.

@AshtakaOOf  I did check all of it still didn't found it

@DevRev What model are you missing? One of the desert mixes?

Nearly a month later, here you go: samle/sd-webui-models/pastel-waifu-diffusion

@Dreamer333 thank you so much!

you're missing the VAE this model is tuned for, which is why the colors are muted. You used to be able to get the VAE here, but it doesn't seem to be on the page anymore and the other links are broken. Trawling other comments, I found this; https://huggingface.co/samle/sd-webui-models/blob/23997884993f591d36bed7c645892d3d0828017f/pastel-waifu-diffusion.vae.pt I haven't tested it, but it's likely the VAE you need. Put it in the same folder as you put your model, and unless you intend to use the VAE for another model as well, just give it the exact same name, and SD should pick up that they're associated. If you still get muted colors after that, there should be something you can find in the webui settings to force SD to use the specified VAE

@oove Thanks, it worked. I had this VAE before, but my SD just didn't use it, so your advice is just on point.