[🔔PETITION] Keep compatibility with the old model please.

[ZeroCool22]

@AUTOMATIC1111

I know everyone is hyped right now with the 2.0 version, but i just see it have a lot of censorship and cut off images from the Data Set.
This mean you will not get "all the possibles results" as you can with the uncensored version 1.4 & 1.5.

And i'm not talking about the Porn aspect only, they removed from the DS all the Artist too.

Here we can see how poorly the results are on female faces/body, it makes very less attractive/stylized.

https://i.ibb.co/hsLPNkc/hvpqrs069v1a1.webp

Hell, even famous ppl are nerfed now, you get poorly results too.

They now show their real intentions to censorship all the future models, so it's in the hand of the community to do and train "Real Free Models" without any type of censorship.

And i also suppose this will transfer when use Dreambooth to train using this "cropped" model.

So i want to ask you to try to keep the compatibility for old models, so we can still be using those in the GUI.

https://www.reddit.com/r/StableDiffusion/comments/z3hlkt/the_stupidity_of_censorship_become_real_with_sd_2/

[Kilvoctu]

I think it's unlikely Auto will toss aside all the models the community built up just for SD2.0.

Also sidenote, feels like I'm the only one who wasn't hyped on 2.0 lol, even before reading about the removed content. I first saw it at #5002 and read the release blog from Stability, and just feeling so lukewarm about what I saw.

[ZeroIQ1024]

My hype lasted about a day. I really liked the idea of the depth2img ... then I saw the results of what can only be described as "Pepe van Gogh" and the hype died lol

[AugmentedRealityCat]

I agree this is bad, but have you seen what you get with Portrait of Jesus by Greg Rutkowski as a prompt ?

[ZeroCool22]

I agree this is bad, but have you seen what you get with Portrait of Jesus by Greg Rutkowski as a prompt ?

[Vega-KH]

That's a pretty epic joke that few people probably get. Because it actually looks like SD 2.0 output!

https://theworld.org/stories/2012-08-25/amateur-restoration-botches-jesus-painting-spain

[ZeroCool22]

I agree this is bad, but have you seen what you get with Portrait of Jesus by Greg Rutkowski as a prompt ?

[bobbybrowns]

I'm not going to update, that's for sure. Seems like we're getting a lesser product in many ways. I'm not seeing enough improvements on things I value and lots of downsides.

[some9000]

Wow, those are some really crappy results. Especially since I made an extension with 280 (so far) different artists to mix and match.

I went through hundreds of artist names. If present, an artist has a HUGE impact on the result. If that's removed the artistic element is basically gone. Then it's just cheap stock photo imitations with crappy ghost watermarks all over. During testing when an artist wasn't present then the results were these weird sort-of guesses based on their name (basically a name from certain region will produce results of certain nationality). Which is what we see here.

Hopefully this isn't the way they are moving forward with.

[eadnams22]

I’d suggest trying art styles/eras instead of specific artists. The art seems to still be in the training, just not identified as specifically as before.

[Kilvoctu]

That extension does have art styles/art movements.
For me, I like using both to really tell the AI what I want. "baroque style, by caravaggio" has more profound effect than just the art movement. Pretty effective at getting the desired style without using many tokens.

[AugmentedRealityCat]

You can also come up with creative juxtapositions, such as baroque style by Pieter Cornelis Mondriaan, or Paleolithic rock carvings by Maurits Cornelis Escher.

[some9000]

I’d suggest trying art styles/eras instead of specific artists. The art seems to still be in the training, just not identified as specifically as before.

Like Kilvoctu kindly mentioned the extension does much more. So it's easy to see what exactly affects the results. Feel free to give it a try https://github.com/some9000/StylePile

[cryptoguard]

V2 makes me want to go back to MidJourney.

[PyrateGFXProductions]

Check out the F222 model: https://youtu.be/EZ3ovOiJmPo

[Verszy]

I wouldn't worry about a breaking change like that; we've all built so much around 1.4 and 1.5 up to this point, from custom models (DreamBooth or straight-up resume-trained madness,) the sometimes gargantuan amount of model merges a lot of us have, hypernetworks, and embeddings that may not work right away in 2.0, and enough software for both training and inference that it probably dwarfs the average Steam backlog in cumulative file size. It'd be a non-starter in virtually everyone's eyes out of those who've used even a little of what we already have. This is definitely a good case for relaxing a bit to do things right, at least as relaxed as the SD dev space gets.

I see potential in 2.0, or at least in OpenCLIP and the 768x768 training dataset (however carelessly auto-curated it may be,) but it's currently just a base for third-party finetuning in my eyes. It'll be good to be ready when others start leveraging those parts, but we have some really green grass on this side of the fence. Even if I wasn't feeling a bit sour grapes about 2.0 myself and it just flat-out impressed me, I'd still be way more interested in the stuff that's been/being trained on 1.4/1.5 right now.

[bobbybrowns]

I hope you're right.

[PyrateGFXProductions]

<!-- /* Font Definitions */ @font-face {font-family:"Cambria Math"; panose-1:2 4 5 3 5 4 6 3 2 4;} @font-face {font-family:Calibri; panose-1:2 15 5 2 2 2 4 3 2 4;} @font-face {font-family:"Segoe UI Emoji"; panose-1:2 11 5 2 4 2 4 2 2 3;} /* Style Definitions */ p.MsoNormal, li.MsoNormal, div.MsoNormal {margin:0in; font-size:11.0pt; font-family:"Calibri",sans-serif;} a:link, span.MsoHyperlink {mso-style-priority:99; color:blue; text-decoration:underline;} code {mso-style-priority:99; font-family:"Courier New";} .MsoChpDefault {mso-style-type:export-only;} @page WordSection1 {size:8.5in 11.0in; margin:1.0in 1.0in 1.0in 1.0in;} div.WordSection1 {page:WordSection1;} -->Try the F222 model. Much better results. https://youtu.be/EZ3ovOiJmPo https://ai.zeipher.com/  Sent from Mail for Windows From: EmilioSent: Thursday, November 24, 2022 9:53 PMTo: AUTOMATIC1111/stable-diffusion-webuiCc: Rex Pirata; CommentSubject: Re: [AUTOMATIC1111/stable-diffusion-webui] [🔔PETITION] Keep compatibility with the old model please. (Discussion #5024) I agree this is bad, but have you seen what you get with Portrait of Jesus by Greg Rutkowski as a prompt ? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>