The BEST way to Run Stable Diffusion for FREE!
Vložit
- čas přidán 7. 09. 2022
- Link to the Colab in this video: colab.research.google.com/dri...
Dream Studio: beta.dreamstudio.ai/dream
Grisk GUI: grisk.itch.io/stable-diffusio...
Huggingface Stable Diffusion: huggingface.co/spaces/stabili...
Lexica: lexica.art/
General AI Playlist: • General MattVidPro AI ...
MattVidPro Discord Server: / discord
My Twitter: / mattvidpro
The AI I Use to EDIT VIDEOS! www.descript.com/?lmref=nA4fDg
Second Channel: / @matt_pie
Helpful Minecraft Tutorial Playlists:
Installation Tutorials for Minecraft: • Install Tutorials for ... Minecraft in-game Tutorials: • Minecraft In-Game Tuto...
THE THUMBNAIL USED IN THIS VIDEO INVOLVES THE USE OF AI-GENERATED IMAGERY
Thanks for watching MattVideoProductions! I make all sorts of videos here on CZcams! Technology, Tutorials, and Reviews! Enjoy Your stay here on the MattVideo Productions Channel, and subscribe!
Leave Suggestions, Thoughts or a Nice Comment! Bye! - Věda a technologie
Hey Matt, to use Inpainting, there is a drop down just under 'Render Images' which is next to the button you click on to run. There are three options in that drop down, one of them is Inpainting to set the 'Mode' for Inpainting.
Shoot. I am beeg idiot.
@@MattVidPro Ruined the whole video, j/k...t5hanks for all your work!
thx! immediately came looking for this!
@@MattVidPro fully expecting a part 2 now 😬
@@MattVidPro I tried a test collab the other day with inpainting, and it looks like it used the same interface, which required the base image and a mask file (instead of allowing you to paint the mask directly on a canvas).
If this collab is the best (for the time being) for inpainting, I'm sure people will appreciate a step by step example using another editor to make the mask or whatever, because it's a bit confusing if you're not used to it ("What? I need an image editor to make the mask?")
I have literally watched every video you have made recently looking for any way of inpainting for free and then immediately this video comes out!
This one is really usefull, you must try the pro, its cheap and you have no limits of usage. Im saying I had been making more than 2000 images the last days, you can retry your inpainting how many times you want, but this function is new, they added it just 3 days ago. They update it everyday, sometimes twice. If you want to share some tips and experiments, we can keep talking in this comment.
Im making some kind of stock of images for a free project, I can say its all about probabilities. More images being rendered and more success, some of them are completly useless. I store everything in the drive, and you can make more images in less time if you use "skip preview".
@@SebaBuenoHaceMusiquitaJijiji pro?
@@imhotep_lukata yes, google colab has a pro and a pro+ version, and that increases the speed of rendering quite a lot, and there are no limits of usage, its really cheap, because it has no limits, just the time (monthly payment)
I know it may seem repetitive to you but we are loving the Stable Diffusion coverage! Please keep that coming!
WE LOVE THE CONTENT STABLE DIFFUSION IS TAKEING OVER MY LIFE
We appreciate you Matt....fire content!
I appreciate YOU random citizen!
Thank you for this video so much
Very welcoming. Thank you.
Stable diffusion is so addictive, using the Colab now, quicker than running on my pc, although the interface is better on my pc version.
great vids, thanks
is it better to use the colab site, or get an appropriate card and run it locally?
or is a local install more limited than the colab?
What about video creation as well?
Thanks
Wow this is so awesome. I can't stop generating images.
THANK YOU SIR
Keep up the stable diffusion content! Have you found a free/paid rest api for stable diffusion?
I'll just note that I was able to get it working locally on my M1 Mac with MPS, just a lot slower than using CUDA on my PC. I haven't tried, but I would guess pure CPU on the PC is also possible, just slow. So needing CUDA isn't a hard & fast rule to run locally, but getting it working without it requires a bit of tweaking things and CUDA does make quality of life a lot better.
i got it running on my almost 6 year old laptop somehow. it's not fast (it takes over 1000 seconds to generate a single 512x512 image with 50 steps) but it still works, and that's cool
What was the prompt/seed for the cat-in-armor thumbnail? That looks amazing!
I would love to know that as well. I've tried to create some armor and nothing comes out close to that.
Did you grabbed that Cat from the thumbnail from Reddit? if yes it was my post haha. Happy to see it here.
i like your sd stuff
Hey Matt please make another video using the inpainting ...I've been trying to get it working in colab without success I would love to see you have a crack at it
amazing
good content bro
really nice
I do feel like a A.I overlord this week. How else do you explain dissecting and adding features to your own A.I. to build a awesome personal A.I that can pump out your thoughts. why have someone else build your platform when all the files are here for you to do it yourself. So glad I own a 3080.
This space is moving FAST!
Trying to use this one. Even rejoined Colab Pro and it still won't work at 512x512pxs, only upscaling 1 and nothing else changed. and I'm only getting "out of memory" errors.
There is always a chance someone on your discord, will use their local installed Stable Diffusion to generate images for someone.
I looking basically for 2 things in free AI image-video generators:
The most coherent and accurate output from text input, and the best quality, resolution. Owning these features means that you can create everything.
I just wonder, something which I am looking for, if there is an AI model, app, which can write the text you want on the object you want, setting the font as someone needs. That's what would be useful.
If there would be a text to 3d model generator that would be great too.
Hello.
I can't achieve to get a picture of a man (ideally a knight) carrying on his shoulders a horse with Dall-e. The only results I get are always a horse with a man on ots back, but I am looking for the other way around. Which AI can do correctly what I asked?
Keep ‘em coming.
What do you think of Dream by Wombo? Which is free (with a few limitations) and you can run it anywhere. I’v been using it for a few days and the results are really good.
Made a few videos discussing it, I like it!
I’ve been using Stable Diffusion GUI by cmdr2 and it is fantastic. It is quite large of a file size, but it auto updates every time it runs and does not require any coding or command prompt knowledge. It includes an initial image option as well.
would you give me the uppdate link plz ?
Thank you for this! I'm using it right now and it works wonders once you got it all set up.
Whats ur spec?
@@sidharthan1784 I have pretty good pc, I9 9900k, Rtx 2080ti and 32gb Ram
@@sane5028 thats insane altogether.. what costed you around???
im using the one from Hugging space , the demo one, The Error is not much an issue as usually if you click again right away it works , the results have come out really good ,my profile pic was also created using this 😁, weirdly its taking a bit more time since the new ver 1.5 was launched,not sure if that has affected it in anyway, but overall I'm getting better results ,not bad at all
It takes an eternity to generate an image tho
@@pneumonoultramicroscopicsi4065 true, can tests one patience at times , lol
Im making thousands of images using this colab, Ive been payed for the pro, its amazing! The secret its to choose the right settings, the first tries will be messed up
what would be the right set up plz ?
@@hatuey6326 Not sure, but some of the images I am getting just blows me away. Seems to really love portraits of women.
@@hatuey6326 I will come back with a good answer, thats why I havent responded yet, Im gonma make a short video and upload it to youtube, so Matt can use it if he wants to
Im rendering a lot of images to make a graphic showing the results; different outcomes thanks to different setting, Im working on the CFG scale, 70% of the images I need are already done
@@SebaBuenoHaceMusiquitaJijiji Found a bug with the colab just now. Seed I use doesn't match the seed it says. Reuse the seed and I get the same pic BUT if I use the long one it says I do not get the same picture.
Wow! Great video, just got access to Dall-e 2 (Wasn't impressed) Stable diffusion is way better!
nops stablediffusion colab is great ! its fast to!! and hes always updateing
At this point I will only use local versions on my gpu. It works so much faster and I can monitor who actually gets my images. Or block it fully on my firewall.
I think the PROMPT_FILE expects a JSON format, is that right? What should the format be?
did you check out the Photoshop plugin? from Christian Cantrell
I use a temporary email site and just keep smashing out images in dream studio.
Does the collab not work on mobile? It took 0.2 seconds to finish running the cell, which obviously can't be right. It doesn't seem to work for me.
I know in painting doesn't work for mobile because they say it but img2img too?
You can avoid connecting to your Google account by going about half-way down in the Diffusers Method cell and unchecking “USE_DRIVE_FOR_PICS”. When the images are generated you can still right-click/long-press on them to save to your device.
I tried that, but I still got the Google Drive prompt. It then crashed when I selected "no thanks"
Not sure if anyone os doing this for hugging faces stable difusion that takes 5 min. I just open up 20 tans of stable difusion and copy and paste the sentences in. Time I get to 20 or so the 1st one is usually done.
switch the node to prompt
Where is the link for the first best free way to generate images website?
Will it have diffusion v1.5 soon?
Google colab isn't free either. After a number of itterrations it will prompt you that you have reached a maximum of Cuda calculations. afther wich you can choose to pay a monthly fee or not
where is the picture from the thumbnail?!
is there a reason the colab page doesn't load up well and is slow for me
colab is still better,
as all diffrenrt model link notebooks are already available , according to one's need,.
HuggingFace also has a free version
anyone know how touse Dreambooth on the colab version? or train this on a specific kind of art?
Well google colab is only free for a limited amount of computing. Pretty soon you have to pay for it. What is the best method for completely free service where you don't have to pay after a set amount of time - is that then only the huggingface and GRisk that you can use or are there other/better alternatives?
run it in your pc, but u need powerful gpu
3060 12gb got a lot of hate. But then stable diffusion came along. The extra vram really comes in handy when doing higher res AI outputs. In Grisk, I've been able to create images up to almost 500,000 pixels, before upscaling. 704x704 or 640x768 or 576x832, etc. Other versions of stable diffusion that include features like img2img and GFPGan eat up a little more memory so I have to knock off 64 pixels from either the width or height, otherwise it crashes.
That's my experience anyway. I'd be curious to know what others are getting for maximum resolutions on their cards.
I'm still on my 1080 and I can only do 512x512. Pretty good for it's age though. Steps at 30-50 take about 20-40 seconds per image...not bad for the ability to just crank these images out and test out settings in the background while doing other things.
@@jenkem4464 It seems 512x512 is the highest res for any 8gb card. I think SD is tuned for that size of image so as long as you can process that without crashing, you're doing pretty good.
@@tungstentaco495 Ah ok, good to know. I'm curious though..what kind of it/s (iterations per second) are you getting with the 3060? I get about 1.2-1.4 right now with a 512x512.
@@jenkem4464 at 512x512 I get 3.4 it/s. my 3060 doesn't start getting down to 1.4 it/s until I crank up the resolution to 640x640. 704x704 is just over 1 it/s on my system. I can't go any bigger than that without crashing though.
@@tungstentaco495 Ok great, thanks for the info.
This code is private when I'm using? Because in 20h a message from a guy appeared while I was using it.
Im gonna join the discord and send you a graphic for CFG scale I made, so you can use it on your videos if you want to
I keep getting No module named 'realesrgan' error but Idk what I missed...
Who knows how in NOP's Stable Diffusion Colab uploud an initial image from my pc?
If you have a good gpu and determination installing it locally with the web-ui is amazing.
do you by any chance know if it's possible using a amd gpu? I have a rx 6600 btw.
@@Groovvy81 sorry mate am not sure, check the discord sever?
When will you do minecraft again
How to edit ai generated image....most of all NFT eyes not similar to each other...make video on that issue
Hi, I'd like some help because English is not my first language and I sometimes have trouble understanding non-written text. Until now, I used this program, mostly because I didn't understand Huggingface's fast speech and I didn't understand which files I have to install on my Google Drive, why I have to ask for tokens again and again.
If I understand you correctly, this program - although it doesn't require a Huggingface ID and token - is just as good as Huggingface. Is there any substantive difference between the two in terms of finished production? If so, is there somewhere on the web a Huggingface step by step simple installation guide?
Presumably, it's "NSFW Disabled: NOP's Stable Diffusion Colab v0.42 (1.4 Weights)" will also include the newer verzos? Because I'll stick with it then. Thanks a lot!
There is no quality difference
@@MattVidPro Thanks!
Hello Matt and humans! I made a video talkin about CFG scale, its in english and spanish, showing examples and how to use it. What is the best CFG value? Spoiler: It depends on what you want to create
I have a new idea for a new channel name: SmartLemon AI
I'm just glad I can use my 3080 for something lol
For me there is no field to type the prompt.
free google collab cut you off from the GPUs after using it for like 1-2 hours and u need wait till next day
The Colab page won't work here.
Me who is still waiting for some minecraft video.
True
As a Swede with rather poor English grammar skill I want to know how private and secure it is to run it on Google Colab, for I had made a prompt "A girl fairy with wings sitting on a leaf floating on the surface of a clear blue tropical lake", It made a rather good image but the fairy was large and holding the feaf, so I added the word "small" that I think is just the size, but the AI generated it as a child, so I had to change "girl" to "lady" and that made it just odd, and then after some more prompt I came to the realization that I should use the word "female". But then a thought came, who else can see what I wrote and what images it created, what if someone take that single prompt out of context, how do the collab work, what information are we sharing, and who are we giving access to so they can see or edit whatever we have on our Google Drive?
copy-pasting someone else's comment:
"If you don't want to connect your google drive, you can just uncheck USE_DRIVE_FOR_PICS before running the cell. I kind of like the idea of images saving directly to my google drive, but it asks for wayyy too many permissions. It has permission to literally delete ALL of your files if it wanted to. Lol. This is because they just mount the whole google drive basically as if it were a flash drive (drive.mount('/content/drive')). I get why they did it this way, but I don't have time to sift through all their code to make sure it couldn't fuck up in such a way where it deletes unexpected files. Lol. I'm sure it would probably be ok, but I don't have my google drive files mirrored anywhere else, so I'd rather not risk it unless I've thoroughly reviewed the code"
So this isn't working anymore right? Now u have to pay to use colab GPU in order to run this.... which I found hilarious
Those are sliders, not dials.
If you want to buy a GPU, I would do so right away because, according to the president of Stable Diffusion, there aren't enough GPU in the world to keep up with demand. It's similar to another cryptocurrency mining run, but without the inevitable crash.
M1 and M2 macs can run stable diffusion.
Is this as powerful as midjourney? Your results are a bit disappointing compared to some of the things I've seen from midjourney.
Colab Link is down.
If you don't want to connect your google drive, you can just uncheck USE_DRIVE_FOR_PICS before running the cell. I kind of like the idea of images saving directly to my google drive, but it asks for wayyy too many permissions. It has permission to literally delete ALL of your files if it wanted to. Lol. This is because they just mount the whole google drive basically as if it were a flash drive (drive.mount('/content/drive')). I get why they did it this way, but I don't have time to sift through all their code to make sure it couldn't fuck up in such a way where it deletes unexpected files. Lol. I'm sure it would probably be ok, but I don't have my google drive files mirrored anywhere else, so I'd rather not risk it unless I've thoroughly reviewed the code
Yeah I just use a secondary account for that with an empty Gdrive, but good to know we can still run it without syncing to the drive.
@@jenkem4464 Oh. Not a bad idea. I might end up doing that actually
appears to be broken
I'm getting a bassterd python error. Anyone have an ideas
I just get an error about pipes. Rip lmao
Mode: inpanting
Does this work on mobile?
Yes it’s difficult though
Any mobile suggestions
OMG FIRST
secondetheth.
im already using it local on a AMD ryzen cpu. it takes about 2 minutes to generate a picture and you need to go through that python install hell. you get a webinterface afterwards and its cool. also has basic in paint feature.
but just head over to midjourney and make multiple throw away accounts if you want some fast pics. licensing etc in this state doesn't matter anyway anymore. if ur afraid about it, throw a generated picture into any transformer and shake it up a little. its funny how fast those companies wont make any cash, thanks to stable diffusion.
But we really need a workflow oriented AI designer like photoshop that can be used to really combine everything together for designers. also I'm really missing something that generates vector and transparent based graphics.
I want to draw transparent comic characters and then apply shadows/styles to them. I want an AI that understands, this is my comic character, and i want it to be able to make it rotate its head. I want AI to work together with me. By now i just see it generate super complex stuff somewhat accurate. Like "yeah mr i stole everything from artstation, i want you to help me design a button. no not a 4k cinematic, trending on artstation. button. like not like this. no like omg. ok my button now looks like a human transcend into a higher form"
There are AIs out there that can do some kind of this stuff. But we really need to combine everything together in one designer that is work flow based. And i honestly think getting that software to work will actually be more work then the AI itself lol.
Right now its like a guy screaming RANDOM BEAUTIFUL FREE PICS, BEAUTIFUL FREE PICS, BEAUTIFUL FREE PICS for everyone and it really is just a fancy neural remixer to piss of every artists that ever worked hundreds of hours on real art lol
And.. we dont even know if you can get copyright to the designs or if those companies are even acting legally.
my laptop has a RTX 2080 lol
Or you could just run it yourself on your own gpu
Colab crashed
It's not free at all. It requires a powerful gaming computer. There is no web-based free SD that i've found
Is there one that allows NSFW?
Please give me some Kudos🎉
Literally have to give them login for gmail and google drive lol. yeah F that no thanks. it should just let you type any email addy in a window and sent it there. there's no reason it needs keys to your google drive
I accepted , will I have to change passwords?
You look scary
You're not combing your hair back. Greatly disappointed
Im getting a haircut on Sunday
@@MattVidPro Glad to hear dude
LOL You can be on a modern Radeon GPU but simply because is Radeon the software would rather work with a 1650 lol Absurd
I just don't like Google Colab since it locks you out unexpectedly at some point during the day when you don't have a subscription. Atleast Kaggle is clear about how many hours it gives you per week. Then it gets replenished afterwards.