Upsize Your MidJourney or Dalle-2 AI Artwork Resolution for FREE!
ฝัง
- เผยแพร่เมื่อ 12 มิ.ย. 2024
- There is a fantastic method to not only upsize your AI art from MidJourney, Dalle-2, or Stable Diffusion, but it can also maintain or even add to the details while making the image larger! All of this can be done for free using a specific GAN model (generative adversarial networks) that has been trained on exactly what your images should look like and how to even improve upon them.
Do a shot every time I say "fantastic" as the bonus game. :-)
Here is the link to the GAN models I have used in this video. Select the one that works best with the image you want to upsize. upscale.wiki/wiki/Model_Database
This person does not exist: thispersondoesnotexist.com
The tool I am using is ChaiNNer, and you can download that from this GitHub site:
github.com/JoeyBallentine/cha...
Join my Discord server, where we create ideas together!
/ discord
New! Become a member to get exclusive access to perks!
/ @sedetweiler
Enjoy this video? Consider buying me a coffee! ko-fi.com/sedetweiler
Twitter: / sedetweiler
Instagram: / sedetweiler - ภาพยนตร์และแอนิเมชัน
You sir have blown my mind with this amazing content! TFS!
This is such a useful video! Been on the Midjourney exploration for almost 2 weeks now and wanted to try my lettering works on them, but the image quality/size was an issue. This pretty much solves it! Thank you so much!
Glad it was helpful!
Great, this is just what we needed to be able to get large prints of MJ artworks. Thanks Scott!
You're very welcome!
This is great content and amazingly useful in both my Midjourney and Virtual photography. Thanks for the video, really great stuff. I can't wait to see the results on my work.
Thank you! Glad you found it useful!
I have never heard of this image processing tool before. Thank you so much for sharing
I thought it was pretty ingenious!
Here's a tip, if you generate a 512x512 image you can put it through img2img with the same prompt and set the res to 1024x1024 and it basically upscales the image and add or change details so it's not just super smooth or plain
Yup! Back when I made this video that wasn't really a thing outside of dalle. I have a video on this trick coming soon. Great suggestion and exactly the way to bend these new tools in creative ways! Cheers!
@@sedetweiler what is the current best way as I’m just about to follow your video step by step in November 2022…thanks
@@cellardoor7500 I still do it this way for the most part.
Like others have said this is a lifesaver! I love your channel! Thank you for sharing.
You are so welcome!
lifesaver ? really ? how stupid are you ?
Thanks so much for this vid. I found there were some changes with the program but was able to match the instructions with the new form. I'm enjoying experimenting with the different models. No such thing as too much fun.:-) thanks again
Thank you for introducing me to chaiNNer, it's awesome!
Glad you found it as useful as I did!
want to say thanks for the tidbit of info and keep up the good work !!!
Thanks, will do! Thanks for taking the time to leave a comment as well! Cheers!
I'm currently at the point at which I create fairly high quality compositions with Midjourney, so having this versatile upscaling method is super useful, the only really crucial thing missing in my eyes is a free or somewhat not too expensive alternative to Dall-E 2's capabilities of modifying sections of the image using prompts. If there is one with comparable quality then it'd be the perfect budget high quality MJ trio which I'd likely utilize for work.
I have been using the free Dalle credits where possible, but also my own photography. You might try the site I mention in the video for generating faces and perhaps borrow eyes from there until MidJourney gets this figured out.
@@sedetweiler Borrow eyes? Video tutorial on it will be much appreciated:)
@Gravega All human-made art is a remixed copy of a copy of the things and experiences in the surrounding world anyway. Your point is invalid.
What you're looking for is inpainting\outpainting feature, it is available in InvokeAI which is a UI for stable diffusion models, check out on TH-cam their videos
And the quality depends on the models you use, plus can train your own models using 20-25 images.
Masterpiece tutorial. Thanks a lot. And welcome to the future!!
Glad you like it!
Wow thanks mate, subscribed. This is a very straightforward video for what I need.
Glad to hear it!
@@sedetweiler just wondering but what does tile size option do?
"She needs to see an orthodontist, but other than that..."
This is fantastic, thanks for sharing Scott!
Glad you enjoyed it!
This is lifesaver. Thank you!
Glad it helped!
Your channel is a valuable find for me.
Awesome! That is great to hear! I am glad you found me!
works perfectly, cut my work time down so much, thank you!!!
Thank you Earthling Scott... Mucho appreciated. Now all we need is outpainting batches.
amazing got it working! game changer!!
Thank you so much. Using UltraSharp and Deviance. Insane results.
Glad to hear it!
Very Helpful. Wow. I did now know about these GAN models that are specific to style/subject. I've just been using generic upscalers out there. Have you compared printed work using this technique vs other upres solutions like topaz and others?. Great vid thanks.
Yes, it was a much better upscale than with Topaz or Adobe Camera RAW (by far). Choosing the correct GAN has a lot to do with that. If Adobe added the option to choose the upscaler GAN, they could easily add this to the Photoshop or Camera RAW gigapixel upscaler as well.
Thanks to you, I now can make my crazy dall-e 2 video animation in high res ! wooop wooop !
Wow! Thank you for sharing this, although it is a little bit of techy to follow all these info, you have a new subscriber to your channel.
Man, thank you SO MUCH! This is invaluable info!!! As a veteran designer/Photoshop user, I'm pretty adept at image processing... But since I started experimenting with AI, I've wrestled with exactly this problem - and I have yet to find an ideal solution for upscaling. Hopefully this should do the trick!
This works like a champ for me! I hope you also find it useful.
fantastic and informative video!!!
Thank you!
Very useful, thanks.
You're welcome!
I have never heard of this before. Its works great.
Awesome! Glad it was helpful!
This is great info and I love your channel. Question, should the enlarging be before or after editing in photoshop? I have been making a lot of edits to my AI works in photoshop and didn't know if I should do the edits before of after the enlarge process.
Always before for most things. It's better to go into Photoshop at the largest resolution you can.
thank you for this upscale app
Absolutely!
Cool thank you for the info
No problem 👍
Brilliant tool!
It really is! It opens so many other door as well!
😄🤝 thank you for your times.. We learn a lot from you..
So nice of you to say that! Cheers!
thank you, this was great
Glad you enjoyed it!
Life saver!!
Thanks man 👍🏿
Excellent
Thanks!
Great tutorial on using chaiNNer. Would love some more in depth ways to use it also.
More to come!
Very helpful.
THANK YOU SO MUCH! This is such a useful video! Sorry can you please tell me if it is possible to edit multiple images at the same time in batch with this program? Thanks in advance and thanks for all the always useful information you give us in your channel!
Yes, you can use the batch image loader, but it will take a while.
@@sedetweiler once again I THANK YOU SO MUCH!
Could you do a video evaluating the different models?
One question please, have you any video about how you prepare the color with photoshop? Do you work with the adobe rgb space and then when exporting you pass it to cmyk? Or do you directly work the file with the cmyk space? midjourney images if viewed with cmyk, change a bit, especially more magenta. Thank you for your content.
Thank you sir ♥️♥️♥️
Absolutely! Also, just dropped an update to this where I work on an entire directory at once!
Wow, this tool looks powerful! Much better than any interface I have seen before. So, I can apply complicated logic to process image afterwards or create some workflow in first place. The whole idea around this kind of processing stack is great, but it does lack the image generation node.
I think he is playing with some new ideas, but it is supposed to be a supporting system, not so much one for generation.
@@sedetweiler it looks like by fallowing your channel I will know everything possibly needed to utilizes imagine generation technically. You are discovering really interesting tools and workflows.
@@sedetweiler Why are they even making these programs with such small images? Im sure they assume people will want to make prints. Seems to defeat the purpose.
beautiful, I will use for my narrative paintings, is ChaiNNer works with Mac? thanks
Had no idea about ChaiNNer. Was googling for upscaling because I'm using Dalle 2 to make some NFT's for my LAGOFrame NFT screen. Very powerful looking tool. Thanks!
Glad I could help! It is quite the amazing tool for sure!
How long should a upscale process take? I’ve been running it for 5 minutes now and it’s still not done!
Maybe I will regret this later but I'm using Gimp with Gmic until I can buy Gigapixel AI. That being said I will definitely try this method eventually.
wow - the tech is exploding
It sure is!
Hi! I was wondering how this upscaling would compare to Photoshop. Thanks for any info!!
MJ testp is amazing for faces/people. I hope they keep it around.
Yes, they are and they have already indicated it will get another upgrade tonight or tomorrow if all goes well.
Awesome. Can you make another video with what you can do with Chainer, some other stuff that is useful to you ?
Sure thing!
Your thoughts: With all is this great AI out now or soon, What do you think Adobe will do? Think of all that clip art/stock they rent. And Content Aware fill looks a little 2019:-) They have the $$ to be a major player. Or will they be the Eastman Kodak of Art and design?
Only time will tell. I know they are working on a new diffusion system for Photoshop that is under closed Alpha now and I hope to get my foot in that door soon to try it out. So, they are embracing it.
@@sedetweiler please make a video if you get access!
The face generator is freaking terrifying.
Yeah! I totally agree!
ok thx!!!
I think I got lost at the part where you showed us a wiki, mentioned a specific site that had people's faces on it, and then demonstrated the upscale site that had you selecting from 3 files/images(?) at the 5:03 mark. What exactly are those? Could you be willing to demonstrate a full walk through of what those are, what made them, etc as I'm still confused what happened here
Those are just the GAN models I had downloaded. You can download the one you feel will work best, and I tried a few. There are hundreds on that site!
That image might be my favorite i've seen generated by MJ. Do you have a link for it?
Not yet. I have yet to finish it.
Thanks for the tip with chainner. This tool is great. I have several upscalers in use, but not CodeFormer. Does anyone know if this works and if so, how?
hey thanks this is great, I see it runs on Linux too. 👍
This is amazing. I’ve been searching for a good upscaler for a really long time. But I’m an idiot when it comes to installing software like this. Would you be willing to give a quick tutorial?
It was really pretty simple. I just did the windows exe and downloaded the options offered. the rest of the work flow is all in the video. Give it a try and see if you get stuck. Cheers!
@@sedetweiler Same here! How do I get that 4x GAN you used?
@@sedetweiler Oh! That’s strange. I was looking for the exe file but didn’t see it. I wonder if my antivirus app saw it as a threat and quarantined the file. Thanks for the tip! I’ll try again. BTW… Love your videos!
@@sedetweiler Where can i find the models you're selecting at 5min 16s you sec?
Great video! Is there any way to run a facial restoration/correction model through ChaiNNer (rather than simply upscaling)?
He is working on that and I think it is released, or will be soon.
Thank you for the great tutorial.
Could you, please, advise, how to install ChaiNNer tool?
Sorry, I am not so proficient in GitHub yet...
HI, I am trying to follow along with your video but it seems like ChaiNNEr doesn't have that LOAD Model button in it. How do I get it to have that button to add models?
Go down near the bottom under pytorch. There is a load model node
It looks like theres an option for iterating frames in a video. Can we see a demo of that?
I will have to look into that. Midjourney has that, but it isn't what most people think it is.
What would be the best one for landscape paintings from Midjourney?
Oh, there are so many! It's really your preference.
Thank you! You deserve a like! Unfortunately, I can't use Chainner... Installing for 30 minutes and it's going this is not normal.
I would keep trying, as he makes updates almost daily from feedback and issues. Especially since Colab is going to start charging, this will be nice alternative.
Did everything the same but for some reason it gets super slow and doesn't seem like upscaling the image on the saving part, not sure why its struggling
PyTorch doesn't seem to run on my iMac because it has an AMD graphics card. I'm trying to use ONNX but it doesn't seem to recognise the .chn files
Oh, so it's better than Topaz Gigapixel? Thx for the info!
I think so, especially when it comes l comes to faces.
Dqanke für die Hilfe darf man es benutzen und dann die Bilder noch kommerziell verkaufen nach dem Upscaler ? danke dir
Thanks this tool looks like just what i need but I cannot for the life of me get pytorch installed on my mac?? how did you do it on your pc? I have installed homebrew, python, anaconda, and have installed pytorch and checked it via terminal, but Chainner wont see it or download it for itself :/ (download wheel just spins forever). Any advice much appreciated!
I wish I could help, but I am not a Mac person at all. I would have no idea where to start. :-(
there is a bug in v0.11.1 for macOS, there is already solution (not merged yet). Just wait for new release.
@@FrenzyCZ so can this be used on my IPad Pro? I even downloaded Working Copy app that deals with forking….but I’m stuck how to run the install of chainner and connect it to my working copy app
my computer sucks. can i use cloud computing to do the heavy lifting? like colab.. or hugging face?
What is a good GAN for landscapes?
Hi Scott, I download the new software from ChaiNNer version: 0.11.5, but it get a lots of time to upscale the image. And then I get an error
Same, its basically just crashing my laptop
maybe you do not have a powerfull graphics card
Are these upscalers safe to use if you're going to use your art for commercial stuff? Will you still "own" your AI art, or can these upscale sites steal your art?
The one I use is local.
is there any AI method that takes a photo and makes it look painted like that? so you could still recognize the person? thanks!
Not sure. There probably is, but the next question is if they would make it available. Photoshop and Corel Painter both have something like this now.
Or you could put the photo online and use it as a reference and tell the AI that you want it as an oil painting or however you like. It'll change it though so if you want the exact image then yeah, apply a PS filter.
How can I get this Gans/models from 4:56? I Installed all dependencies but those models just missing... Thank you!
How long does it usually take? I have a decent pc but the image im upscalling is still not finish. I’m only trying 1 image as a test but 30mins later it is still not finished.
Depends on the size you started at and where you are going.
Does someone know how to upscale 2 times automaticly, instead of starting the process again?
Now I’m here trying to see if I can trick my iPad into running any of these :p Any insight on mobile or browser run GANs/Upscalers?
You can run some of them in Google Colab, and that is web based so you can probably get it to work from your iPad. It isn't gonna be fun though!
@@sedetweiler Appreciate the reply and your videos. I did find a nice stable diffusion ai run on Collab, ran into my monthly limits in under an hour :p I suspect a dedicated upscaller wouldn't be *as* GPU heavy.
Im using the 4xUltraSharp model and I keep getting the "cuDNN error: CUDNN_STATUS_NOT_INITIALIZED" error message .What am I doing wrong??
Is it possible to do a batch processing in this application?
Yup!
Everything works except for the output part, the image never went there, no error messages either. Any idea what might have happened?
i have the same problem
Hi! the same problem! Have you found a way to solve it?
How do you even download chainner on Mac? Says it can’t be verified
No idea. I don't own a Mac and know little about them. This system is supposed to work on a Mac from what I understand.
Hello! Many ThankS for your video! I've downloaded the v0.12.4 but I can't find the node "Load Model" could you pleas help? Thanks
It should be down with the other PyTorch items.
I downloaded Working Copy app from apple store for my iPad Pro, will I still be able to utilise Chainner on the iPad or only only a pic?
Not sure, I don't use an ipad.
So i tryed to do this but its just continues to spin and never finnish try with a few images, do you know what the fix is for this
it can take a while, especially if you have other things open that are using your GPU. I do these overnight and batch them.
Hmm, it doesn't seem to work for me. Getting stuck on exporting, any ideas?
It takes a long long time. Close as ll other apps. I let mine batch and run overnight.
Thanks for the Video. Unfortunately chaiNNer will not save any image for me? The program runs and there is no error showing, but the directory remains empty after it finished. Does anybody have experience with this issue?
I have the same issue on my macbook pro
Why do I need these GAN models to upscale my images? Can't I do it without it? Also, do you use today a different method to upscale your images?
Just Upscaling is fine, but the gan will add details and fix the image as it scales it. Eyelashes for example, where a normal upscale won't see those details as they were probably just 1 or 2 pixels before.
hi, is it possible to upscale more image at once ??
yup. I pass it into the upsizer twice.
My Mac won't launch Chainner. '"“chainner” cannot be opened because the developer cannot be verified."' Thinks it could be malware. It's a bummer. Maybe that'll get fixed in the future.
This is great... How do you install it?
Just go to the page where it has the distributions and click on the one you need.
You are the best!!
aww! thank you!
Aahhh node based! Sweeet.
I know! I was super excited to see that!
I feel I am doing something wrong here, I keep getting this error "An error occurred in a Upscale Image node:
A VRAM out-of-memory error has occurred. Please try using a more extreme tiling mode.
Input values (partial):
• Image: RGB image 2048x1152
• Number of Tiles: 4"
No matter what I have for the tiles it fails, have you seen this before?
That is a pretty big image. You have a nice video card with a lot of vram?
@@sedetweiler hah I realised I too my RTX5000 out and it has a crappy K5200 in it. It's dual CPU so have been churning them out via CPU.. slower but not in any hurry! Thanks for answering.
Thanks for the great information. I have tried to follow your instructions but I get this error message in chaiNNer -
An error occurred in a PyTorch Upscale Image node:
cuDNN error: CUDNN_STATUS_NOT_INITIALIZED
Input values (partial):
• Image: RGB image 1024x1024
Do you have any ideas what the problem is?
Solved my issue - changed from GPU to CPU and error message gone.
Oh, that is good to know!
Where would I get the .pth file for Gigapixel?
anyone know of a pytorch alternative for AMD GPUs?
Is there a CLI or Python API for it?
I went through over 250 images from this person does not exist and found the same issue with everyone the eyes are not correct; either they are too large, too round or just not quite the right shape.
Saying that the pictures are all outstanding.
I still think it's pretty impressive overall.