Stable Diffusion 3 Release Date Announced.
HTML-код
- Опубликовано: 2 июн 2024
- Preview SD3 with Stable Artisan Discord bot stability.ai/stable-artisan
Get early access to videos and help me, support me on Patreon / sebastiankamph
Chat with me in our community discord: / discord
Stable Diffusion for Beginners Playlist • Stable Diffusion Begin...
My Weekly AI Art Challenges • Let's AI Paint - Weekl...
My Stable diffusion workflow to Perfect Images • Revealing my Workflow ...
ControlNet tutorial and install guide • NEW ControlNet for Sta...
Famous Scenes Remade by ControlNet AI • Famous Scenes Remade b...
![Gunna - back in the a [Official Video]](http://i.ytimg.com/vi/oBrbejOHjLw/mqdefault.jpg)
2:28 : The lack of NSFW support meant the community chose not to support it and most people don't want a censored model. People want enough optionality to be able to turn those branches on/off or include/exclude, they don't want the entire branch lobbed off because someone else decided what should and what shouldn't be censored because that's how you end up with a disaster like Gemini.
exactly. if i cant lewd my favorite waifus what's the point. has there been any word about this in regards to SD3?
Not just NSFW, but also other restrictions, like inability to match to a named prompt, forcing you to use Loras instead to do that.
@@2099EK Yeah
You can make plenty of nsfw pictures with sdxl. Maybe not the base model but who cares?
@@jonathaningram8157 You missed the important part and got fixated on the NSFW part.
Because you dropped the date in the front of the video, I in turn will not now click away but let your video run to the end. This is called "fairness" to those of you trying to hustle past all reciprocal relationships towards god knows what.
this single gesture made me like, comment and subscribe!
Yeah, people being respectful of my finite life span get respect in turn. Instant like. Science papers have abstracts for this reason.
up
Thankyou!
God bless Stability AI. I hope they manage to overcome current difficulties and not go bankrupt or sold.
Thanks for not delaying the answer. Staying for the whole video.
That's very kind of you, thank you :)
Going to wait until there are NSFW fine tuned checkpoints available
will we be able to finally have a girl on all fours without looking like a monster?
I'd love to see how well the Creative upscaler works. Its a shame they charge so much for it 25 tokens. So the basic sub gives you max 36 upscale for $9.
It is only important for me that it is free. As always and what has been in the past. 😁😁
I didn´t believe it would come out.
Much LOVE to the peeps at Stable Diffusion !!
Man, I love you videos. They are simple and funny
That's so very kind, thank you 😊🌟
Thanks, as always, appreciate your content and the additional dad jokes to torment my family with 🤣. Glad to hear it will be dropping soon, can't wait to have a play.
Sebastian with the news!😊
😁🌟
SD 2.0 had a more restrictive license. I won't use it for my commercial projects when sdxl exists.
You are a real one for dropping the date at the beginning. Subscribed and liked just for that
Thanks!
Wow, thank you! Very kind 😍
@@sebastiankamph I had to thank you some how. Every tutorial i searched for you had done.
Thank you for the video, because I didn't see that I had received the email. This can happen when you have several emails, including one reserved especially for the AI stuffs.
Oh cool, not long to wait. I won't hold out much hope for hands but it would be nice if faces could be fixed. One of the big reasons I never stuck with some other UI's, was because they didn't have anything like Adetailer (at that time). Which made them useless as I mainly deal with faces.
I look forward to your video on it.
Yeah, it's actually a hassle to keep using adetailer just to fix something as simple as faces on anything further away.
Thank you so much for not wasting time and for giving us the information first thing. I stuck around to watch the rest of the video because you did this.
So nice of you, thank you! It really helps :) 😊
Does that mean this release is uncensored? Or I guess the way it works people will release checkpoints based on it that will be uncensored?
I used the service and cancelled it. It was extremely confusing how to use what service. I couldn't figure out the video diffusion.
Wuhuu can't to try
Make Mine Marvel? No. Make Mine Kamph!
😊🌟
i lold
Thanks for sharing the date. Of course I watched until the end after that. Thanks for the great content and the jokes.
Thank you, that's very kind of you :) 🌟
Will SD3 be usable on 8gb VRAM GPUs or will it require more?
I do believe that will be possible with some optimizations.
The 2 billion parameters model, doubt the bigger one (8b) would.
so is it not being released open source?
You will be able to download the files freely from Huggingface, but their licensing is "non-commercial" use
What they mean by "non commercial" is that you can't use the base model for profit, you need to contact them to get a license, however I believe that the images made using SD3 can be used for commercial purposes
@@volcanowater Thank you for explaining.
The business model could be created around access to the GPUs of participants ....like bittorrent?
what about animals? most of the time elephants and larger african animals are really hard to get them to look right. copilot does it right immediately. have these been improved?
faller mellan stolarna :D jag som troddde att du var tysk (måste ha varit efternamnet som lurade mig). Iallafall, tack för alla SD videos, har hjälpt mig mycket. Och pappaskämten framförallt =)
Kul du gillar dem! Jag tänkte att dialekten gjorde det, men tydligen inte då 😊
They have not fixed prompt adherence, that is very sad ... without it what's the point of it when there is already a cascade ...
Any idea how long before we’ll have something like a1111 for SD3?
It's a model, so right out of the gate. You'd just need to install and load the model in the UI. Now, if you're talking about custom fine-tuned models based off of SD3, that'll take longer.
@@peterlewis2178 Thanks for the reply!
if i buy rtx 3060 12gb, is it good for generation in sd? and 3?
If you can run SDXL (3.3B) you can run SD3 medium (2B)
I predicted that Stable Diffusion 2 was going to flop. I You can't produce NSFW pictures with it. The only reason Stable Diffusion flourished in the midst of Midjourney popularity was because of the NSFW. Based on my experience with the Civitai website, I would have to say that at least three fifths of the pictures on there are NSFW. With just Pony Diffusion alone, which is a great-looking model by the way, it's probably around 80% NSFW.
By some of your answers, nice to know it will be free for non-commercial use. I don't want to pay while learning how it works lol. I'm wondering if 8GB of VRAM will be enough for that one. Oh well, we'll all know that the 12th of ... (I don't want to spoil the date for the others, just watch the video ^^).
people are able to use sdxl on 8gb of vram so it will most likely be possible unless I'm missing something because the sd3 medium has less parameters than sdxl.
I wonder if SD3 huge (8 billions parameters) would still run on a 3090/4090. Probably not.
Will it be available on automatic 1111 ?
Yes, in due time. Probably first in comfy and comfy-based uis.
funny, i just googled SD3 release date and boom, 46 minutes ago you posted this video :D
I got you covered! 😅🌟
I'm quite excited by SD3 small because with similar parameters as SD1.5 people will finally ditch that model and retrain on sd3 with modest training cost.
It would be great if it can replace 1.5, yes! Imagine having better prompt understanding, higher resolution and more with the power 1.5 has.
So a credit is exactly a cent at the lowest subscription. Given that I've on average generated over 3k images a month, costs would add up very fast to the point of buying multiple 4090s. Though admittedly, I wouldn't try nearly as much random stuff if hitting the 'generate 4 images' button cost me a quarter every time. As is, it costs me maybe 0.2 cents in electricity.
Also, what even is a 'creative upscale'? I like to make myself wallpapers for my computer and with a very recent monitor upgrade, upscaling to 5120x2160 is nontrivial. Only webui forge can even hit that resolution without resorting to the ultimate SD upscale script.
Do you think SD 3 will be better than Midjourney in terms of creating models (women)? 😅
Probably, since 1.5 already has a higher ceiling than MJ (with more advanced work).
Non-commercial only, sigh...
I must echo what others have said: thank you for telling us the date right at the start, much appreciated. Video liked, and watched through to the end! 👍
Very kind, thank you :)
Vram Minimum?
The medium model (2B) will probably need about 6 or 8GB Vram, but until release it's just a guess.
Amazing dad joke
They tried and made a good one, but your dad jokes are still the best XD
Happy to hear it 😁🌟
Only 2B model, - 8B one will be for 💵💵💵😜
What a shame….
@@DaniDani-zb4wd it was expected, but the question is: who will be paying for censored SFW version???
8B weights will release. Stabilityai said that after the release of 2b one, other parameters will release after it including 8b
@@volcanowater They said the weights will release when each model is done training. Most of the demos we've seen until now (and the API) have been the 8B model, but Lykon has said that it is severely undertrained. 2B obviously trained a lot faster and was probably a higher priority due to its greater reach potential. I'm still looking forward to what 8B can do when it is done.
@@NotThatOlivia And guess what the 2b model is just horrible... a disaster.Who would have thought🤣🤣. They spent all the time delaying the model censoring it and not improving it at all
Dude, you get news letters before they are relesed in public web site.
sd3 for 6.5 credits? wow i think stability AI need investor
Wow! They done it wright, and don't hide SD3 behind a paywall. GG Stability AI!
Cost ?
Free ?!!
The local weights will be free for non-commercial use. So majority of users will be able to just download it from huggingface
they should have done a fondation and stayed open source and become like wikimedia, untouchable
I don'tt think there are enough people interested in that
@@jonathaningram8157 at the beginning of wikipedia only a few cared too, stable diffusion is here to stay if it's open source, or it will just become a photoshop subscription in the future
No assistants......
just the model......
Bring back Emad, Robin etc....
Nvidia should finance this project
Have to see how pozzed the assistant is, a censored AI is utterly useless. 😑
So they took the greedy path ? That's kinda disappointing, I guess i'll stick with PDXL and adetailer then
Not exactly greedy, the images made using SD3 can be used for commercial purposes, but you can't sell the model as a service, you need a special license from them. And also they need to take the greedy path because recently stabilityai have fallen into hard times and the company is nearly bankrupt
"Greedy" Broke and unemployed donkey opinions.
I hate the stale "Dad jokes" 😅😂😊
You misspelled "love"
@@sebastiankamph I do love them I often send my sister your jokes. Be yourself man!!
@@sebastiankamph I really love them (and this comment too XD), your videos always put a happy smile on me. Please, never change! ^_^
yes, yes, we all received an update email about the release date. no need to make an 8min video about it
Thank you so much. ❤ I stopped watching the video because of your advice. You are so smart, I'm glad you could help people like me. Sweet bro!
@@brodanschuter4146 :*
Then there was no need for you to watch it, was there? Also, I didn't get any e-mail so yes there was a need.
@@Elwaves2925 hey, you do you champ if you want to spend 8 minutes on 5 secs update news what ever
@@rhym8882 And if you want to waste your time watching a video, that is clearly about something you already knew, then fair play to you too.
#timbanxua Thank you🎉
Just commenting to say thanks for cutting to the news straight away and not dragging it out to the end of a 40min video 🫶
Happy to help 😊🌟