Comments

[hidden]

So I feel like I should probably comment on this, as we've been exchanging messages with @adelinesinclair4 for the last week or two.

First, let's make something perfectly clear - you're not upset about the loli/shota. You've been here and seen those posts for a while, if you thought they were that disturbing, you would've at least switched that setting back to default so as not to see them on a daily basis. But you messaged me today with a screenshot of a bunch of recent loli posts, so that's not the case. Not only that, at one point you literally told me (and I'm sorry that I have to quote a private conversation, but it's just very relevant here):

My argument here isn't even "Hey this website shouldn't have CP on it" My argument is "I shouldn't be sitting here considering posting CP to compete with everyone else, because the bar is sooo much lower for it"

So no, it's not the lolis. Your primary grievance is that your posts are not getting approved. I tried telling you multiple times to just spend a few minutes fixing them, because they're getting rejected for specific reasons (bad fingers, etc.) but you didn't wanna hear it. Instead, you were focused on how you're being treated unfairly and that surely this must be a conspiracy by the NHK/CERN. Eventually you decided to request a replacement for a couple of your posts, but those were just the ones with obnoxious big banners; in other words, all you did was post the original images without the added text - you couldn't be bothered to actually sit down and fix any issues with the images themselves, so you just chose the option that required no additional work.

I don't know what to tell you - you seem convinced that your posts are very good and should be approved. When people don't share that sentiment, you throw a tantrum and start looking for excuses not to fix them. With that attitude you're not gonna get far, I'm afraid.

Updated by strong hand

[hidden]

noppersnop said:

It says img2img in the meta data, so we dont know how much of it is classic digital art and how much AI, but yes this is very nice and creative , very hard to do without controlnet

it says img2img when you use "enchant" function that is basically just sd upscale.

this pic is generated using t2i

[hidden]

Ralem said:

Thanks, I got my start yesterday watching a video from early 2023 and didnโ€™t realize this field moves at light speed. I recently got better models that donโ€™t have this problem and make much better art. Iโ€™ll post some of the stuff I made tomorrow. Thank you for the reply.

Good luck my young padawan.

[hidden]

ElReyBlanco said:

Oh boy, there's a long road ahead for you, full of delicious knowledge.
In short: it's a variational autoencoder. In this context, it's a thing that enhances the generated picture. In this case, it's gonna make it so picture won't look desaturated.
As for how to use it - it depends on what are you using to generate pics.

Thanks, I got my start yesterday watching a video from early 2023 and didnโ€™t realize this field moves at light speed. I recently got better models that donโ€™t have this problem and make much better art. Iโ€™ll post some of the stuff I made tomorrow. Thank you for the reply.

[hidden]

Ralem said:

What is that? Literally just started yesterday and was using this model a tutorial told me. Can you explain that a little more? Whenever I used this model it always seems to come out like that. Itโ€™s my favorite model when it comes to the actual art but the coloring looks kinda bad. Hereโ€™s the model I was using btw:
https://civitai.com/models/135440/breakdomain

Oh boy, there's a long road ahead for you, full of delicious knowledge.
In short: it's a variational autoencoder. In this context, it's a thing that enhances the generated picture. In this case, it's gonna make it so picture won't look desaturated.
As for how to use it - it depends on what are you using to generate pics.

[hidden]

Ralem said:

Bro how do you even go about making this? How would I get this pose with a different character using img2img? Sorry, complete noob who just started here. Also stunningly beautiful image.

It says img2img in the meta data, so we dont know how much of it is classic digital art and how much AI, but yes this is very nice and creative , very hard to do without controlnet

[hidden]

Sirius_fw said:

@Ralem,
It doesn't matter what model.

It's just that some models have VEA built in, others don't.
Without VAE they produce dim colors, some do not work at all.

If you use WebUI by A1111 or WebUI Forge, then:
Just download some VAEs, put them in the .\models\VAE folder and select them in the user interface.

I recommend using these:
https://huggingface.co/SiriusArchive/VAE/tree/main
Which SDXL use for SDXL, which is different, use for SD1.5.

P.S.
In fact, just don't bother whether there is VAE inside the model or not, just use VAE manually, just in case.
For SDXL this is not important at all, all VAEs there give almost the same result.
For SD1.5 it depends more on VAE, but "perfect color VAE" usually gives a good result, I settled on it when I used SD1.5.

P.P.S.
Also, for anime models you almost always need to set the โ€œClip skipโ€ parameter to 2.
This has nothing to do with your colors problem, just saying in case you don't know.
Almost all anime models work better with "Clip skip" 2.

Thanks bro. And yeah I saw the clip skip in the tutorial. I think that I had a shitty model. Gonna try using the newer ones and if they have that issue Iโ€™ll download some of those. Appreciate it bro.

[hidden]

Ralem said:

How to fix it? Sorry Iโ€™m brand new and was just completely following a tutorial. Using break domain v10 I believe for all of this style. Hereโ€™s the model I was using btw. https://civitai.com/models/135440/breakdomain
Just bad model?

@Ralem,
It doesn't matter what model.

It's just that some models have VEA built in, others don't.
Without VAE they produce dim colors, some do not work at all.

If you use WebUI by A1111 or WebUI Forge, then:
Just download some VAEs, put them in the .\models\VAE folder and select them in the user interface.

I recommend using these:
https://huggingface.co/SiriusArchive/VAE/tree/main
Which SDXL use for SDXL, which is different, use for SD1.5.

P.S.
In fact, just don't bother whether there is VAE inside the model or not, just use VAE manually, just in case.
For SDXL this is not important at all, all VAEs there give almost the same result.
For SD1.5 it depends more on VAE, but "perfect color VAE" usually gives a good result, I settled on it when I used SD1.5.

P.P.S.
Also, for anime models you almost always need to set the โ€œClip skipโ€ parameter to 2.
This has nothing to do with your colors problem, just saying in case you don't know.
Almost all anime models work better with "Clip skip" 2.

Updated by Sirius fw

[hidden]

ElReyBlanco said:

Seems desaturated, have you tried using VAE?

What is that? Literally just started yesterday and was using this model a tutorial told me. Can you explain that a little more? Whenever I used this model it always seems to come out like that. Itโ€™s my favorite model when it comes to the actual art but the coloring looks kinda bad. Hereโ€™s the model I was using btw:
https://civitai.com/models/135440/breakdomain

Updated by undeleted 43937

1 28 29 30 31 32 33 34 35 36