MidJourney In Trouble? Open-Source and FREE AI Generator SDXL Is Stunning (Comparison)

Подписаться 270 тыс.

Просмотров 148 тыс.

50% 1

In this video, we look at the new version of Stable Diffusion, SDXL 0.9, which is a huge leap forward in quality for open-source AI art generators. This puts it on par with MidJourney, but it's free, nearly unlimited, and completely open-source. Plus, it can be run on most modern computers locally. I do a direct comparison, what do you think about the new SDXL?
Enjoy :)
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Links:
Blog Post - stability-ai.squarespace.com/...
Github Repo - github.com/Stability-AI/gener...
Clipdrop - clipdrop.co/

Наука

Опубликовано:

10 июн 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 191

@chariots8x230 11 месяцев назад

8:01 The ‘relight’ feature is really impressive. It’s very useful to be able to change the lighting of the image generated in real time.

@matthew_berman 11 месяцев назад

Agreed

@digidope 11 месяцев назад

Thats post processing and not anyomore AI image generation. Photoshop has the same plugin. It's actually probably same open source code in both.

@Dr.W.Krueger 10 месяцев назад

Nothing new. We had machine learning based relighting in our compositing pipeline since 2018. The only thing that got better over time was execution speed.

@benjamininkorea7016 10 месяцев назад

Wait. . . like faces and stuff? Can I inpaint my own face, and get a new lighting angle?

@benjamininkorea7016 10 месяцев назад

@@digidope Does Photoshop have it currently? I've had a 3 year-old version of Photoshop installed because the 3D lighting effects were canceled.

@STG113 11 месяцев назад

With the right mods, prompting techniques and good models, SD older engines can already get images as good as midjourney

@sppedben6095 10 месяцев назад

Nope, not yet. The main thing that baffles is the resolution. Even if you upscale the highest, hands are better with new models. Not even talking about creativity. And idk what version you are talking about. 1.5 is aged, 2.1 was a huge flop. Looking for sdxl 1.0, and hopefully it can challenge midjourney.

@Maisonier 11 месяцев назад

Certainly! We would greatly appreciate it if you could create a tutorial for installing this Stable Diffusion version. It would be even more amazing if you could provide an introductory guide to all the tools associated with Stable Diffusion, such as Dreambooth. This guide could cover topics like model training techniques and the benefits of utilizing LORA for different models. Additionally, including best practices for organizing folders, managing Python versions, virtual environments, and avoiding conflicts with other tools like Oobabooga would be incredibly helpful. We would be thrilled to have access to such comprehensive resources, and we want to express our sincere gratitude in advance. Liked, already subscribed and shared.

@yazmo109 11 месяцев назад

agree, should explain how install it on any drive and change the models path as well

@matthew_berman 11 месяцев назад

Guide for SDXL locally coming soon

@Maisonier 11 месяцев назад

@@matthew_berman you're the best of the best

@River-Kosmos 9 месяцев назад

Is it out yet? @@matthew_berman

@pseudonym9667 11 месяцев назад

This is not a fair comparison. Featured images on mid journey are ones that happened to turn out really well with that particular prompt, it's the one in a thousand best result. Run it again and the results aren't nearly as good. So you're comparing the cherry picked one in a thousand best results on mid journey, with one shot results on stable diffusion. This is an unfair comparison. You need to regenerate the image on mid journey with the prompt, and then compare that one shot result with the one shot result on stable diffusion. That is a fair comparison.

@UncleUncleRj 10 месяцев назад

Yeah, my images never turn out very well with these despite multiple different prompts and many regenerations.

@WillDanceAlone2U 10 месяцев назад

Great video. I'll try it.

@digidope 11 месяцев назад

Midjourney seems to release new features every week. Last week we got zoom out and some orher things and last night they released --weird parameter. In 4 weeks it might be quite different what it is today as MJ6 is expected to come out.

@Accou25 11 месяцев назад

this is my "favorite tv show" for the last few months

@pavellegkodymov4295 11 месяцев назад

Same to me :) quite useful and practical. Thanks to Matthew for great work

@matthew_berman 11 месяцев назад

Thank you :)

@juanchogarzonmiranda 11 месяцев назад

SD has always been better than MJ, the plasticity and tools it offers are superior, as well as being free, now that SD generates high quality images tools like ControlNet will give wings to creativity and professional production.

@flcamera 10 месяцев назад

Thank you for the info, Midjoury kinda confuses me and wanted to generate instead of grab b-roll for my next short (as I did with AI story but Crayon and Dall-e weren't giving the imagines I wanted as I wanted some a closer to "real" look) so this was of great help. Just signed up and used up all my tokes :)

@johnblack9499 11 месяцев назад

Great video, spent a lot of time playing with this yesterday. I struggle to get the images to be super realistic however the quality is amazing!

@matthew_berman 11 месяцев назад

I feel like it does realism quite well. Keep playing with it and hopefully you can get it to do what you want!

@johnblack9499 11 месяцев назад

I'd only been trying to do animals up until yesterday, now I've tried some people and it does do humans much more realistically :-) thanks for the encouragement to keep giving it a go and keep up the great work with the videos!

@JonMurray 10 месяцев назад

This is absolutely insane. I love it. Awesome video man. New subscriber ✌🏻

@jpfister85 11 месяцев назад

Awesome! Can't wait for you to do an install walkthrough once they release the full model, because my computer can actually run this locally!

@matthew_berman 11 месяцев назад

Coming soon!

@kakhak 10 месяцев назад

lol

@pastoryoda2789 10 месяцев назад

I hope i can install and run smoothly without my laptop burning 🔥

@Jessy-Noel 9 месяцев назад

The best option! I spend hours trying to come up with good coloring pages with other AI app... I decided to try with SDXL 1.0 as my last option, and it really blew my mind!!! It's really incredible!

@TheDailyMemesShow 10 месяцев назад

Thank you for posting this information, just subscribed 👍💪😊

@darkman237 11 месяцев назад

Can you use these created images in a commercial product? If so what documentation is needed?

@matthew_berman 11 месяцев назад

It's open source, so I assume you can, but I'm not 100% sure.

@gileneusz 11 месяцев назад

8:30 this is soooooo coool.

@matthew_berman 11 месяцев назад

Agreed!

@4cryinoutloud291 10 месяцев назад

Great video! Are we able to upload a photo to give the AI a starting point?

@calliopelove 10 месяцев назад

What about privacy options? Are all images public or can we do anything privately?

@bazookaman1353 11 месяцев назад

Does anyone know if I can run on 6GB vram even if super slow? My SD 1.5 runs with some limitations for comparison btw.

@thebecoming6740 10 месяцев назад

How to take off the logo “stability a.i “ on the lower right corner when I want to download a photo

@vovanchik_ru4208 11 месяцев назад

When will there be a video on the fine-tuning model?

@matthew_berman 11 месяцев назад

Fine-tuning LLM? I know i know...i've been delaying that too long.

@dunknow9486 11 месяцев назад

Thank for this info. Please give update for this project progress

@mbrochh82 11 месяцев назад

This is pretty exciting indeed. Crazy times.

@allenstroud4236 11 месяцев назад

I agree. I've spent at least two hours and it was great! Thanks for sharing.

@featherly4267 10 месяцев назад

Is it generating in desirable ratio type??

@cocoy 10 месяцев назад

Excellent review, thanks!

@Fittrendz 10 месяцев назад

Do we need pro to get rid of watermark?

@crazydaysss2546 10 месяцев назад

Great video sir and great news from clipdrop 🎉

@matthew_berman 10 месяцев назад

Thanks!

@thenext9537 11 месяцев назад

Interesting, my prompt MJ scripts I have them nearly perfected. I started using things like --seed and --no (with a long exclusion list). This is VERY interesting though. I have a 2080 ti, so I think I owe myself a run at this. Those using API, you can start using API calls to get your MJ game up too. The tools still are mixed though, gpt api calls sometimes go sideways.

@jopansmark 11 месяцев назад

Wow, all that Midjourney users control is seed and negative prompt and they're happy about it

@aarobriggs2665 11 месяцев назад

I cant wait to put my hands on that thing ❤❤

@holdthetruthhostage 11 месяцев назад

When 1.0 hits it's over and now Meta vision is out and GPT4 VISION eventually we won't see issues with creation any more

@ALFTHADRADDAD 11 месяцев назад

Meta Vision?? Is that image recognition?

@LouisGedo 11 месяцев назад

8:14 Wow........incredible feature!

@matthew_berman 11 месяцев назад

Agreed

@rogerbruce2896 10 месяцев назад

Please create a video on how to get this up and running on our local computers? one test I do is ask these generators to draw someone on a unicycle juggling clubs. No generator including SDXL gets it right. Someday perhaps :)

@darkman237 11 месяцев назад

Did you use the photo realistic preset?

@matthew_berman 11 месяцев назад

I did not.

@darkman237 11 месяцев назад

@@matthew_berman It makes a huge difference! Also, don't be fooled as I was, the 3d model preset doesn't give you a 3d model, just the look of one.

@angelevangelista7506 10 месяцев назад

Very interesting! But is there a Mac version??

@Wanderer2035 11 месяцев назад

Can we use it on Mac?

@jayhem4624 11 месяцев назад

I admit I'm impressed with SDXL, it's really good at following prompts and much easier to use than Midjourney. Midjourney is still better, especially for power users (I have over 30K images on MJ), but SD is catching up fast! Can't wait for SDXL 1.0, especially if they fix their tools, as the 'uncrop' tool is not working at all, it can barely handle backgrounds but nothing too detailed, whereas MJ 'Zoom Out' works wonders.

@matthew_berman 11 месяцев назад

Yeah, I’m super impressed. Especially because it’s free.

@jopansmark 11 месяцев назад

SD is catching up? It always was a better choice and the best solution on the market. Also "Zoom out" is stolen from Automatic1111

@ratbassterds 10 месяцев назад

Wow thanks. I used to pay for mid journey and these other AI art aren't the best. But this is up there and has other great features. 👍

@yapyh2872 11 месяцев назад

Please do a video on how to install it locally, Thank you for sharing.

@matthew_berman 11 месяцев назад

I definitely will when it’s ready

@bettyeverett8089 10 месяцев назад

Sounds exciting. Will definitely give it go.

@patricklogan6089 11 месяцев назад

Another great vid!

@musicandhappinessbyjo795 11 месяцев назад

When is the local install coming out?

@HaroldCrews 11 месяцев назад

Mid-July

@4cryinoutloud291 10 месяцев назад

So I can't run the program on a M2 Mac book?

@Otchengazoom 11 месяцев назад

Fantastic! Thanks.

@dunknow9486 11 месяцев назад

Hello, can you a video on how to get consistence with Diffusion?

@DavZell 10 месяцев назад

The log at 1:45 doesnt look like a log because its a rock outcropping. 😂

@sarder2008 10 месяцев назад

Yes every time you ask for an image creation you need to wait at least 5 minutes, some free service !!!!

@alwilson6471 11 месяцев назад

Could not install - cloning repository failed due to authenticity issues - Permission denied then could not read from repo.

@matthew_berman 11 месяцев назад

That’s something specific to you you’ll most likely need to fix your GitHub keys

@alwilson6471 11 месяцев назад

@@matthew_berman Sounds like a plan :-). Thanks

@allchatgptprompt 10 месяцев назад

This sounds like a game-changer! 😲🚀 MidJourney facing challenges, but the introduction of an open-source and free AI generator like SDXL is a truly stunning development. Embracing open-source technology can foster collaboration and innovation within the AI community. Excited to witness how SDXL will empower users and revolutionize AI generation. Kudos to the team behind this remarkable initiative! 🙌💡 #OpenSourceAI #SDXL #AIInnovation

@SpikeyWombat 11 месяцев назад

Will it work with amd video cards?

@matthew_berman 11 месяцев назад

I’m not sure

@ALFTHADRADDAD 11 месяцев назад

Fucking nuts, i did not know about the relight feature

@matthew_berman 11 месяцев назад

So cool right?

@TheNerdiestKids 11 месяцев назад

If you're going to mention Midjourney, you need to mention it together with Niji, which has more illustrative imagination. Mid journey ain't close to in trouble with this. It's an okay start tho.

@matthew_berman 11 месяцев назад

never heard of Niji...

@TheNerdiestKids 11 месяцев назад

@@matthew_berman Midjourney is in partner with a Japanese AI company called Niji. WHen putting in commands post prompt, if you add "--v niji" it will use that version. It's far superior for comic book and super creative anime type imagry.

@digidope 11 месяцев назад

@@matthew_bermanNiji is the "secret" behind all those MJ images of women with less clothes. If you add --niji it seems to ease the moderation filter. And while it's in theory a anime model, it seems to do quite photorealistic stuff too.

@jopansmark 11 месяцев назад

@@digidopewhy ease moderation on MJ if you can just use SD without moderation at all?

@digidope 11 месяцев назад

@@jopansmark It takes a lot more effort to get same result from SD than from MJ. Some think it's cool when you need to write long prompts and long negative prompts, and select right model, with right LoRa's. For most people that is too much. I personally prefer simple method, but do use SD for work as i use my own models and LoRa's.

@commander_fu6457 10 месяцев назад

400 Images for Free*. I also read they are watermarked. So its basically a free test version and not actually free.

@PandorasBox2 11 месяцев назад

dang thats amazing

@sumitmamoria 11 месяцев назад

pretty cool

@rockumk 11 месяцев назад

What level of modern GPU?

@luminousdragon 10 месяцев назад

Base SD XL is not onpar with midjourney, to me that seems very clear. But when people make loras and checkpoints for it specifically and they are fairly tuned itll be significantly better than current midjourney. THough we dont know what midjourney will be capable of then But as always, SD has WAY WAY more flexibility if you take advantage of the various extensions and tools.

@johnguild8850 10 месяцев назад

Wow. Very impressive.

@Juan2142 10 месяцев назад

It's funny how all these videos on youtube always have to mention Midjourney in their title 🤣 Then when doing the comparison through prompting it turns out that it looks nothing like Midjourney 😂 I would say its better to just talk about an AI model with doing a comparison instead of trying to compare it all the time to Midjourney, because lets be honest MJ is in a class of its own.

@tlothompson6935 9 месяцев назад

You're crazy. SDXL not only is capable of amazing life like pictures, it also has models to where you can get the same looking person in every photo with ease. All for no monthly subscription and private ($50 a month for MJ). I haven't used MJ in a few months, due to the pricing of course, but I don't see why you'd need something "better" than SDXL.

@Juan2142 9 месяцев назад

@@tlothompson6935 🤣you do realise the irony of your comment right? 🤣🤣🤣🤣🤣🤣

@raylovecars 10 месяцев назад

For More Realism you should select style -> Photography

@dunyazade 10 месяцев назад

So a few more years and we'll all have our own AI...?

@PhilKingstonByron 10 месяцев назад

Very COOL

@BatsiraiMusuka 11 месяцев назад

I’m blown away by the ReLight feature. Time for me to play around with AI again (l had made a departure from Graphic Design work for a bit…busy with other things). 😅

@kakhak 10 месяцев назад

Who the f k interested what you quit and what you do now?

@Quik_Dev 10 месяцев назад

Did it have an API

@digitalphanes206 10 месяцев назад

1:48 the log doesnt look super real, says pointing at a rock...

@muazashraf409 10 месяцев назад

Can we sell these images as NFT?

@kasforai 10 месяцев назад

Free is watermarked?

@financialchimes4546 10 месяцев назад

I tried it, but it still has problems with hands.

@DallePics 10 месяцев назад

Will this work on AMD cards or NVIDIA only like most self-hosted AI art generators?

@21stcenturyscots 10 месяцев назад

Thanks for not posting a link

@digidope 11 месяцев назад

Maybe it would be better to wait that the model is actually released before doing clickbait tittle? The Clipdrop demo of SDXL0.9 is not at MJ5 level and MJ is now 5.2. When and if the final model is released MJ might already be in V6. Its funny how many youtubers have same title for a video that is based on a company press release. As a long time daily user of Stable Diffusion, i would take everything that Stability AI promises with grain of salt. Emad is known not to be very trustworhy. He is a sales man after all. I really do hope they deliver all what they promise, but let's see when it actually happens.

@matthew_berman 11 месяцев назад

This is a working and great alternative to MJ. I don’t think there’s anything click-bait-y about my title.

@StewieCrypto 11 месяцев назад

@@matthew_berman I assume digidope is calling you out for the "MidJourney In Trouble?" bit of your title as clickbait.

@digidope 11 месяцев назад

@@matthew_berman Midjourney is not in trouble as 99% of MJ users don't know what Stable Diffusion even is. Stable Diffusion just got really bad news: Many medias, including local newspaper here in Finland is writing about AI generated child porn. And they mention Stable Diffusion as the tool. It's just a matter of time when politicians are talking about it. So it's bad timing for Stability AI to come out touting XL as open source AI tool that can do anything. Politicians are known to over react on things.

@jopansmark 11 месяцев назад

Yeah, SDXL v0.9 is not at MJ5 level, it's above

@major-skull-gaming 11 месяцев назад

looks like its going to be great im waiting for my first image to be done right now :) thanks for the info ;)

@major-skull-gaming 11 месяцев назад

holly crap this is amazing i love it ! relight is crazy cool but allllll the tools on the site are so great too

@luisdanielacevedo7378 11 месяцев назад

This is even more lighter than Stable Diff?

@Leviathan_070 11 месяцев назад

this video could have been just a link, I can read what he is reading, nothing new here... pf

@Dieucreateuruniversel 10 месяцев назад

for free, and at time code 5:27 we can see price...

@Vetter07 11 месяцев назад

I feel like I saw this same exact video on someone else's page...

@matthew_berman 11 месяцев назад

Nope it’s mine :)

@bawest4 10 месяцев назад

love it!!!

@fredflintstoner596 10 месяцев назад

Mrs Richards: "I paid for a room with a view !" Basil: (pointing to the lovely view) "That is Torquay, Madam ." Mrs Richards: "It's not good enough!" Basil: "May I ask what you were expecting to see out of a Torquay hotel bedroom window ? Sydney Opera House, perhaps? the Hanging Gardens of Babylon? Herds of wildebeest sweeping majestically past?..." Mrs Richards: "Don't be silly! I expect to be able to see the sea!" Basil: "You can see the sea, it's over there between the land and the sky." Mrs Richards: "I'm not satisfied. But I shall stay. But I expect a reduction." Basil: "Why?! Because Krakatoa's not erupting at the moment ?"

@hardkur 10 месяцев назад

i tried XL 0.9 yesterday thru automatic1111 and it was super garbage , most likely i did something wrong (images from XL 0.9 made on the website are quite nice)

@EcoAncestralAmorico 10 месяцев назад

i went to the web of sdxl and is not free, just some try images and then you got to buy credits !

@BR-hi6yt 10 месяцев назад

Not many human-like AIs - just aliens and weird things.

@l.hartmann950 10 месяцев назад

All I get are bizarre images.

@matthew_berman 10 месяцев назад

Like what?

@l.hartmann950 10 месяцев назад

@@matthew_berman i inputted "izzers tennis score keeper", my product, and there were two-hands on the ends of arms... melted rackets...

@samitdigitalart 11 месяцев назад

Time to say goodbye to midjourney. They don't even offer a free ride nowadays

@FrankSchwarzfree 11 месяцев назад

dang

@edgar-mmxxiii 10 месяцев назад

The reason I subscribe to Midjourney for $60 a month is its stealth image generation. Does SDXL offers the ability to hide our images from public? If not, never mind.

@Lurulf_ 10 месяцев назад

if you run it locally on your computer with Comfy UI or Automatic 1111 its hidden, unless you share them yourself :)

@tlothompson6935 9 месяцев назад

Ya, this was the exact reason I stopped using midjourney. I only wanted it if it was private and of course they have that feature locked behind the highest paywall. SDXL is private though as you are running it off your own computer.

@LouisGedo 11 месяцев назад

👋

@chariots8x230 11 месяцев назад

Can it run on Mac OS?

@chariots8x230 11 месяцев назад

Now that I see the system requirements, it looks like it will only run on Windows or Linux. Bummer 😟

@cybervigilante 10 месяцев назад

It can't do simple editing of uploaded images. I just wanted a dark blotch removed from some forehead skin and it couldn't understand what I wanted.

@lawrencehawkins7198 10 месяцев назад

Problem. SDXL wants, nay demands, access to your email, damn them. Be sure to cook up one without your vital data and contacts.

@Rikmenda 10 месяцев назад

With censorship included... 😒

@MarkDemarest 11 месяцев назад

#STUD 💪🔥🙌

@Bartetmedia 11 месяцев назад

It gives you free credits and $9 a month afterward. Great tools, especially the Uncrop. Now I can fix cut-off heads... yaee!!

@matthew_berman 11 месяцев назад

🔥

@AlessandroCalzaStudio 10 месяцев назад

No midjourney is not in trouble because results on SDXL 0.9 are worse the Midjourney V4

@jopadjr 11 месяцев назад

897th...Thanks Matthew

@jamesyoung7400 11 месяцев назад

Good luck getting 'approved' for access to it.

@matthew_berman 11 месяцев назад

What do you mean?

@radioreactivity3561 11 месяцев назад

The model is coming in mid-july.

@markcelis_eyepopdesigngrap9988 10 месяцев назад

So, just to be clear: this AI will NOT run on Mac OSX?! -------------------------------- (you know: the OS that MOST artists and other creatives use!?)

@Lupusteppae 10 месяцев назад

Hope this time my comment won't be deleted...I wrote: nice but you forgot to mention what about Mac users, when you compare SD with MJ...Not all the world use Windows!!

@hugboat808 9 месяцев назад

AI generated images will simply become a tool of a human who will still need to be a skill artist to use these tools. The lazy prompt model will become what it already is, trivial app traps.

@PaulBohno 10 месяцев назад

400 watermarked images a day, isnt really 400 free images a day.

@StewieCrypto 11 месяцев назад

5:32 - MidJourney has a trial period for new users so you can indeed test it quite extensively for free.

@yapyh2872 11 месяцев назад

In the past, yes. But that's no longer the case, every attempt to test it will be met with 'Due to extreme demand we can't provide a free trial right now...'

@jopansmark 11 месяцев назад

Nope.