ULTIMATE QR Guide! Don't Skip These Steps.
Vložit
- čas přidán 16. 06. 2024
- Check out video sponsor www.thinkdiffusion.com as we create scannable QR Art inside Stable diffusion with ControlNet.
QR Code Generator: keremerkan.net/qr-code-and-2d...
Download brightness model: huggingface.co/ioclab/ioc-con...
Install ControlNet: • How to use ControlNet....
Install Stable Diffusion: • How to Install Stable ...
Also check out new QR ControlNet models: huggingface.co/DionTimmer/con...
FREE Prompt styles here:
/ sebs-hilis-79649068
Support me on Patreon to get access to unique perks! / sebastiankamph
Chat with me in our community discord: / discord
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...
LIVE Pose in Stable Diffusion • LIVE Pose in Stable Di...
Control Lights in Stable Diffusion • Control Light in AI Im...
Ultimate Stable diffusion guide • Stable diffusion tutor...
Inpainting Tutorial - Stable Diffusion • Inpainting Tutorial - ...
The Rise of AI Art: A Creative Revolution • The Rise of AI Art - A...
7 Secrets to writing with ChatGPT (Don't tell your boss!) • 7 Secrets in ChatGPT (...
Ultimate Animation guide in Stable diffusion • Stable diffusion anima...
Dreambooth tutorial for Stable diffusion • Dreambooth tutorial fo...
5 tricks you're not using in Stable diffusion • Top 5 Stable diffusion...
Avoid these 7 mistakes in Stable diffusion • Don't make these 7 mis...
How to ChatGPT. ChatGPT explained in 1 minute • How to ChatGPT? Chat G...
This is Adobe Firefly. AI For Professionals • This Is Adobe Firefly....
Adobe Firefly Tutorial • Adobe Firefly Tutorial...
ChatGPT Playlist • ChatGPT
CHAPTERS
0:00 QR Intro
0:51 Deep Dive
3:42 Create QR Code
5:44 Workflow 1
21:26 Workflow 2
FREE Prompt styles here www.patreon.com/posts/sebs-hilis-79649068
Check out Video sponsor www.thinkdiffusion.com
Hi! Does ThinkDifffusion A1111 instances support --api ?
workflow 1:
tile - 0.5 ctrl, step 0.1-0.8 range
brightness - 0.35 ctrl, full step
depth - full ctrl, 0.2 step
(play with ranges depending on checkpoint/image you are generating)
I appreciate you slowing down and sharing your perspective on how knowledge is being transferred in the space - feeling inspired to deep dive and learn controlnet on a more fundemental level!
Thank you! It took a little longer than usual, but I felt it was important to try to show people the why.
Crazy how the denoising strength has so much influence on the scan thanks for sharing!
You are so welcome!
Thank you for sharing this, I’ve seen ThinkDiffusion mentioned on a few subreddits and I’m excited to apply this knowledge and make my own!
Great video. I have been fighting with this for a while and you were able to help solve some problems and let me understand why the problems where problems. The mark of a great teacher.
One note I would add is that a lower error correction on a QR code means a smaller code, with larger individual color blocks. Functionally error checking makes a more stable easier to scan code. But aesthetically a smaller code can give you a less disjointed picture. The truly amazing QR art you've seen were done with very short urls, and low error checking.
Great input! Most truly amazing QR art I've seen sadly didn't scan on my phone though 😥
This is exceptional and frankly, all videos, opinions, teachings and thoughts of Sebastian are awesome. Many can learn from you. Thanks for this. Much appreciated!
Wow, thank you! Very kind and I appreciate it 😊🌟
Of all the tutorials I've tried this is the only one that worked for me. Thanks a lot!!!
Äntligen en guide som utgår från det självklara!
Well i tried the guides, tried your workflow and it worked much better, i guess the title IS true.
Thanks, I spent a lot of time researching this!
absolute legend, I've been waiting for this, will save to watch once I've finished my current project
Hope you enjoy it! And good luck with the current project 🌟
Really amazing research & comparison on the controlnet and SD settings. It would really help me. ❤ Amazing
Glad it was helpful! Thank you for the kind words 😊🌟
Wow, this is so detailed! Thank you so much for the video 😊
Happy to help! I saw so much weird stuff out there so just wanted to add my take on it.
Thank you very much for your finding! I have recently conducted some QR code tests on control networks, including brightness, tiling, and illumination. I'd like to share the findings of my research.
Firstly, let's talk about the brightness control network. I found that the start value and weight have a significant impact on both scannability and creativity. Based on my experiments, I recommend setting the start value between 0.25 and 0.4, with a weight of 0.4 or higher. Higher weight values improve scannability accuracy but reduce creativity. On the other hand, higher start values enhance the image's creativity but slightly decrease scannability precision.
Next is the tiling control network. I discovered that setting the weight between 0.95 and 1 and the start value between 0.2 and 0.4 yielded favorable results. Higher start values improve scannability precision but lower creativity, while higher weight values enhance scannability accuracy with a minor impact on creativity.
Lastly, the illumination control network. Based on my experiments, I suggest a weight of 0.4 and a start value between 0.4 and 0.45, with an end value of 0.9. I found that excessively high weight values not only have minimal impact on accuracy but also reduce image creativity and result in cluttered output. Higher start values slightly improve scannability, and I recommend setting the end value at 0.9.
thanks for sharing! will try it in deforum 🤔
Fantastic Tutorial. I was wondering how these were done, but I was waiting for you to tackle it. I knew it would work if you did it. Quick question, with using Think Diffusion's service, can you upload a created model to the service? I tested the service and it's great. Just need to upload one of mine.
thank you for the amazing guide JerAx of Stable Diffusion... was waiting for your guide for QR codes.
Happy to help, glad you're enjoying the content! 🌟
As someone who created one of the first videos on generating QR codes, even I can say that this is hands down a better workflow 🎉
Happy to hear it. There's a lot of information to sift through out there! 🌟
Nice work! Thank u
Hi Sebastián, this is very useful as I struggled too to make a readable QR code. One thing is I thought you spent a little too much time about what the other guides recommended. This guide is a little too long and because of that. I know some other youtubers will take ypur guide and make it shorter and to the point....and they won't give you credits! 😕
Congrats on your new sponsor!
Thank you so much for your help!
Thank you, you're probably right! It is what it is. You will know though! 😊🌟
Brightness+ tiles is actually what someone recommended to me a few days ago, so I guess it makes sense
Thanks for your video. Do you have any tip to do a qr code with another image than the qrcode in imgtoimg, like using a logo from an existing project?
realy good overview i found out that all by my self too but 1 litle misstake for the depth part the depth processor does not see black tiles as dark colors when you pre preproc you will see what i mean the deepth looks in a "horizontal ancle and cant look "down" easy or see dark tiles. what would help much better would be a high contrast modle that understands there are always 2 colors that need to be seperated but i didnt found one fiting so far.
"film photography" and "soft focus" from your prompt both contain the idea of "round". All photography terms trend toward a camera lens, and it's especially pronounced in rev animated. (Some "round" lens words that come to mind: exposure, iso, depth of field, focal length. I still use them all the time though. They do improve quality.)
Good input!
Thank you very much!
Usually the others QR-codes are like just two image merged by Photoshop. Not to much connection between the image and the QR-code. In your version I could make a forest shape like a QR-code. And it is working!
And the best, that finally I understand what I am doing! :)
Just the 3th ControlNet, the depth is not working for me. It make me less create QR-codes, so I don't use it.
PS. I love you Stable Diffusion videos! They are so pro but still I can understand! Thanks!
You're welcome! Happy that you got images that you liked from this :)
Thanks for the video!
Is there a way to put a logo in the middle ?
Hey Sabastian, I love your content. Do you think that you can do a video on training a LoRA with the Kohya_SS GUI?
Love this 🙌
Nice, keep it up.
Was able to get this to work with only 2 control net - tiles (no pre-processor) and brightness. Using depth totally ruined and changed images for the worse, no matter the settings.
Tiles = 0.55, .28, .7 and Brightness settings were .35, 0, 1 = success!
HI Sebastian great video, I have a question for some reason ( I am using ThinkDiffusion) the controlnet QR code pictures are not showing when I generate the image and I selected Enable what could be the problem? None of the two ways you showed worked for me because the pictures on the controlnet are not being used to generate the image, is there anything I need to active?
Amazing tutorial! I had a very high scan rate with the ones I created following your steps. I have a question: is there a way to use my own images to do this? Such as in a shot of the inside of a store and use it to generate the QR code with it?
Hi, really curious on how can we do that ! If you have found something please share !
Thanks a bunch !
Hi @sebastiankamph, great video, A little thing to point out the link for the brightness control net model is missing the Config file .yaml.
I totally agree that if the preprocessor change doesn't change how to processed image looks, it cannot matter for the model.
Great insight by Sebastian here! Definitely learned something. Thx,
Thanks for sharing 😊
My pleasure 😊
I hadn't looked into this stuff yet; I had assumed someone had figured out some sort of hack to incorporate QR code readability evaluation in the scoring loop of the the AI to steer it away from deteriorating the code too much...
Where can I find the depth model? Having trouble with this, keep getting errors about yamls, the brightness model doesn’t have a yaml or config
Thanks for this very well detailed tutorial! is it possible to incorporate a QR code in a ready made image on photoshop? i will try your second workflow but in case it doesnt work, your workflow adjustment tip is appreciated! cheers
The thumbnail was created by taking an image and adding the QR code on top, then using it in img2img second workflow. It drastically changed the image, but the pose of the woman remained.
Did you try illumination model for ctrlnet? i thought you could get some color in it, since QRcode works is just differentiate the brightness. so maybe we can add darker color to certain things and the QRcode can works, I didn't try this QRcode thingy atm, just beginning to research a bit on how it works.
great content, ill try the first method later. have a nice day.
I was able to get good results even with the other people's settings. Not every image is scannable, but a good 80%
Happy to hear that! 🌟
Subbed, I looked so hard for good qr content. I have been messing around and made a few but they couldn't scan, I had an idea to upscale the image and when I upscaled to 2k,4k it actually scanned but it still didn't work for some. So yeah I'd love to know what I am doing thank you.
Glad I could help! I felt the same way, couldn't scan all those other beautiful images everywhere.
will have to see the video again . its so detailed. but thanks . even a non technical person like me could understand something . just that my qr codes arent scanning . none of them. anyone can help?
Small tip, make an introduction with all the pitfalls instead of a monologue every 2 steps
what would the route be for upscaling? just ultimate sd with 0 denoise or can we use highresfix/ native upscales etc?
Is there a good way to add or mashup an existing image to the qr code, everything I tried with img2img didn't work, either it's unscannable or the image is completely different. Would love to see a tutorial about that :)
That's all well and good, but I've seen cool QR only once - that is, code that looks like a pretty picture, but it scans like QR code!
Awesome tutorial and the best one yet. Thank you for nailing it just like you did with Tile upscaling, and so many others
You're very welcome! Thank you for the encouragement 😊🌟
What does the tile controlnet do when you're not upsizing an image, does it just increase sharpness?
Awesome!!
Oh, why thank you my green friend!
Hello Sebastian! Many thanks for the video, especially the part on pre-processor, I learned something new. I would like to make a different question for you: do you actually think that there is business opportunity for this kind of QR codes? Are companies/business units interested in paying for having such QR codes in your opinion?
Tough question. I'm sure there are usecases where businesses would want good looking QR codes. Is it a whole business model? Maybe not? I'm not going to discourage you if you feel strongly about it 😅
quick tip.... you might actually wanna set the QR code padding lower in some cases.... when your doing all this... if you can burn 30% of the QR.... you can use INPAINTING Except mode... and keep some critical parts.... like a face or specific object of focus.... by setting padding lower.... you will have more of that. space that you can remove.....
Interesting input and a cool tip!
Awesome tutorial and great idea. Thanks Seb!
You're very welcome, and thank you!
A great tutorial
fantastic stuff! and great timing! does anyone know how viable it is to print these as stickers? I.e if I have a 512x512 image and I want to print it on a postcard or business card - then 2048x2048 should be enough?
Should be enough for print yeah. Digital images are usually 72 dpi, you'd want to get them to 300 best case, 150 minimum. So if you have a 2048x2048 in 72 and scale it down to fit your postcard it should raise the dpi to levels that work well for print. Source: my day job.
i am totally new to this sebastian can u help me out, i am unable to get the qr code incorporated into the output at all, its like the output is only generated from prompts
I found something very interesting:
Why do we use this method for QR-code only?
I made some test. I made black and white images in Photoshop (but could be any other program like Paint), and I used it like the QR-code. I could get amazing controlled results:
- a garden where the flowers make a short writing (a name of a girl)
- a forest shape of a country
- coints that make shape like a person
etc...
So it a very new dimension!
But it will not work with all type of shape or all type of prompts!!
Thank you this is amazing, but so much micro configs
Yes i am one of those ppl who ctrl c ctrl v and have absolutely no idea what i am doing 😂😂😂
Haha, as long as you're happy, keep doing what you're doing 😅
Great video! I dont have 3 uits in Controlnet, is there a trick? Best regards
Settings -> ControlNet -> Drag the slider for more
Can we upload our own image in the workflow 2 ?
Can you explain how can we use it ?
Try mixing this with my logo to art tutorial. Let me know how it goes
Can I somehow save those CN Settings to save some time?
Nice
Time 14:11 - How did you always get the same image? Every time I generate, they gave me different images. Thanks
how can we use our own image though? like own photo or company logo. any idea? adding one more controlnet?
Hi, really curious on how can we do that ! If you have found something please share !
Thanks a bunch !
Hello excuse me I have a stupid issue: I have only one unit in my control net. How do I add the other ones? Thank you for the incredibly great content!
Settings -> ControlNet -> Drag the slider for more
I followed the steps but I dont see my QR code appearing in the image I generate Anybody can help me figure out where I messed up please? Thanks in advance
Nice video! Not sure where to ask this but I’m having a major problem. For some reason my VAEs stopped working in img2img generation, seem to work just fine in text2text, but here they simply aren’t applying at the end of the preview generation, and I see no error log. Nobody so far was helpful in solving this problem. This is killing me. Pls help?
I wish I could help but haven't encountered this. I'd try a fresh install.
@@sebastiankamph
Understandable, thanks anyway. I havent seen this issue anywhere else either. Can I reinstall A1111 without having to redownload all my loras and models?
@@sebastiankamph
Okay so I got some progress on my error I think, I checked the boot up menu and it gives me at the start this error 'Couldn't find VAE named klF8Anime2VAE_klF8Anime2VAE.ckpt; using None instead"
This is exactly why I support your content, you are not a follower! Love your videos!! Thank you 🙏 for everything!
Check it, your presence here is a national treasure, you hidden gem! 💎
@@sebastiankamph 🙌🙌
That joke needed a bigger buildup, but the foundations are there.
Oooh, nice one, building on top of the groundwork I laid! 💯
so basically workflow 1 is like blending the qr code into an environment, and workflow 2 suits layering something within the qr code
I have a question; perhaps you have the answer. I have 10GB of "Dedicated GPU memory" and 32GB of "Shared GPU memory." How can I utilize the second option, Shared GPU memory? Is it possible to do so?
Like if you tried Sebastian's QR codes taken from the video 😅
But in 10:50 you got a black square on the image then preprocessing with inpaint global harmonious corrects it. That means that it does something even if we don't notice it. I think that at the least it replaces transparency with white
ThinkDiffusion looks interesting but I never sign up for a website that won't disclose pricing up front. If I have to sign up (and give them my email) before I can even see prices, then I shy away.
I can see pricing without logging in
I was able to see pricing too 🧐
@@johnnguyen562 Yeah, well it's behind the "Try Now" which I admit I just assumed would be like most every other website ask me to create an account. However I was wrong and you are correct. If I click that button it then tells you the pricing.
I'd still suggest they adopt a "Pricing" page which has become standard so they don't have more people like me who don't immediately see any information on pricing and just leave.
I just want to make a cool image with the QR, but not a really working QR.
Haha, well you're free to do so 😊
@@sebastiankamph I tried but all I got was really awful. For some reason, after being close to the correct parameters, everything just turn into trash... really frustrating.
@@DJVARAO It's really finicky for sure. You can try the last workflow with an image applied together with the QR code in img2img instead of just the QR code.
@@sebastiankamph Great tip! I will do it. Thanks!😁
There's a WebUI extension to directly create QR codes
i tried SD on a mac too slow, SD on Colab keeps disconnecting, SD on paperspace could work fine with experience, SD on thinkdiffusion works instantly quick, reliable and affordable. really fun. thx buddy...
RunDiffusion has been around for like 7 months.
@DaHolby Competition is good, it will make both ThinkDiffusion and RunDiffusion better by competing with each other
brightness doesn't work at all TT
What is the use for this? Did I miss some fad of putting QR codes on top of images? Seems all it does is make for a heavily obscured image?
It's the current fad, yes 😂
Love your Vids however your Discord Mods leave a lot to be desired.
BTW your explanation of why preprocessing doesnt always matter makes complete sense
Thank you! The community mods do hard work and have to deal with a lot of very weird stuff. Sometimes it can feel extra harsh or targeted, but it's never personal and they're just trying to make it a safe space for everyone, erring on the side of caution. Now I'm not saying anything or anyone is perfect, so if you feel you have been wronged, feel free to dm them or me.
@@sebastiankamph I cant as @Rune kicked me (same UN as here) when I suggested that he DM me if he had an issue with what I posted (rather than just randoml;y deleteing without a comment)
. Anyways I've grabbed my coat and moved on
Not about your tutorial, but about the user impetus to do this, continually doing what every one else is doing is the real 'doing it wrong'. Ape chase trend tiktok cringe. You can do this with barcodes, text, encoded messages, silhouettes, matrice diagrams and blueprints. Why focus on QR codes? Ape chase trend.
ThinkDiffusion? So like a green version of RunDiffusion? lol
it's fine to delete the comment Kamph, it wont change the truth.
show me on the doll where the QR code touched you :D
The what now? I haven't even been home. I never delete. CZcams auto-deletes spam or links though.
@@RuneW0lf I like you Rune, and I got no bad things to say about you. Your moderation skills are prompt and you are a reliable mod, which is a blessing to have.
Though you might want to glance over the way you like to tease people and troll them, for self amusement, these are instigator traits which I know you are not.
@@sebastiankamph it's alright Sebastian, it's just Rune being Rune
@@sebastiankamphCZcams deletes pretty much anything. I've written long comments on some videos in some other channels, those go to memory hole it seems. Sometimes two comments in a row and the last one gets removed. Anything codelike or like web links can get deleted. Seems like it has gotten worse in recent years.
have you seen "DionTimmer/controlnet_qrcode" models? "QR Code Conditioned ControlNet Models for Stable Diffusion 1.5 and 2.1"
These ControlNet models have been trained on a large dataset of 150,000 QR code + QR code artwork couples. They provide a solid foundation for generating QR code-based artwork that is aesthetically pleasing, while still maintaining the integral QR code shape.
Oooh, that's wild, great tip! 🌟
I'm releasing a video on this today!
@@RussellKlimas What are your thoughts on it compared to other workflows?
@@sebastiankamph WAY easier. Though I still prefer img2img over txt2img. Only need that controlnet 1.4-1.7 weight is all that is needed. Video is out.
Would love to see how you make it better!
@@RussellKlimas Skimmed through it. Looks good, but I couldn't get your examples to scan with my phone (didn't want to comment negatively on your video 😅)
Clickbait. No they aren"t.
Thanks, don’t wanna waste my time.
Watch the freaking video mate, step by step comparison
I try to only clickbait when I can back it up. This time, I'll say most guides are using a lot in error.