ComfyUI : Hiding words in your images + fixing faces. QR Code ControlNet Pareidolia Hybrid Image

The comfyUI tutorial for today is a fun one! We will hide the word "Shop" in our image, and you probably didn't notice it unless you squint at the thumbnail for this video. This effect is known as a hybrid image, but might also be coined as pareidolia, but that isn't exactly true. Not only will be play with this control net, normally used for QR codes in our images, but we will also add in a chain of nodes to fix any wonky faces that get generated during this process. This is a nice addition to your AI Art adventure! #comfyui #stablediffusion #comfy #qrcode #controlnet #pareidolia
You can grab the QR CodeMonster from HuggingFace, and there is also a 1.5 version as well.
huggingface.co/DionTimmer/con...
You can use the Manager node in Comfy to get the impact pack, or you can use "git clone" while in your custom_nodes directory with this URL: github.com/ltdrdata/ComfyUI-I...
Interested in the finished graph and in supporting the channel? Come on over to the dark side! :-)
/ @sedetweiler

[@sedetweiler]

Gang, I forgot to mention the graph is posted in the Community area on KZbin for channel Sponsors, like always. Thank you again for all of your support!🍻

[@DemonPlasma]

Do the supporters at the $0.99 get access or is it only 4.99 and up?

[@BigSnoots]

Greetings from WFB Scott! Saw your livestream this last weekend and was happy to hear we're neighbors! Keep up the great work!

[@sedetweiler]

Awesome! Thank you!

[@d4n87]

Interesting applied like this, using face detailer as well! I hope the version for SDXL will come soon though, the generations made with that model are on another level 😁

[@sedetweiler]

Same here!

[@jasonsutis3387]

@@sedetweilerbrightness is already released all we need is tiling and it’s go

[@kpr2]

Re: Signatures, watermarks, etc. - LOL! Yeah, a lot of my recent generations are signed LORA at the bottom, or it finds a way to work LORA into the background billboards somehow. Never fails to make me laugh.

[@jackbrittain3771]

Thanks!

[@sedetweiler]

Thank you! ❤️

[@Vestu]

New Detweiler video wooot!

[@sedetweiler]

woot! ;-)

[@saintkamus14]

I did everything as in the video (did have to install the detailerpipe) but, i get an error with the facedetailerpipe =| I've updated everything same results. Any ideas? this is the error: "Error occurred when executing FaceDetailerPipe: Could not run 'torchvision: :nms' with arguments from the 'CUDA' backend." etc, etc.

[@slipely]

I just stumbled upon your channel and its awesome ive started using comfyui. To help with my free lancing work, could you make a video on how to imprint images on to packages ,boxes , bread, chocolate bars? I work with different clients and sometimes i get a backlog of need to do quick presentation pieces with the clients logo, branding tags etc. This is really interesting i am already going through all your videos and playlists 👍 Thank you !

[@sedetweiler]

Great suggestion! I will see what I can dream up!

[@jepperisum]

Great idea! because I am desperately looking for at workflow like this too ! :)

[@MisterBoothCoUk]

Great idea 👍

[@Satscape]

About 5 minutes in I'm thinking....this is inspired by the film "They Live"! 😁 Is there a way to get the face detailer to do a limited amount of faces? I had a room full of people, but only really needed the first two rows of them detailing, not the guy right at the back who's not even facing forward.

[@sedetweiler]

Yes, but that's a bit more complicated. We can have segs index the faces, or you can filter by the bbox size. You can also try disabling the inpaint option.

[@Satscape]

Thanks! I think changing the bbox size may work for this.

[@sedetweiler]

Perfect!

[@brandonflores4]

im getting a error: #### It seems that models and clips are mixed and interconnected between SDXL Base, SDXL Refiner, SD1.x, and SD2.x. Please verify. ####

[@user-wp1lk3qm6j]

i have problem error occurred when executing Ksampler mat1 and mat2 shapes cannot be multiplied (154x768 adn 1024x320)

[@Istarax]

Are you sure you have the correct clips connected? Not using 2.0 clip for xl, or so.

[@jayronnbulworth1915]

My last message was deleted, i guess because i put a link for the controlnet model, i'm sorry i didn't know that wasn't allowed. What i was saying was that i believe you are using the wrong controlnet model, you should try to find one that has the exact same name you see in the video.

[@user-wp1lk3qm6j]

@@jayronnbulworth1915 okay thank , im use control net in description

[@gw-j]

Running into an issue trying to add the Ultralytics control at 7:20. I have the impact pack installed and updated, the ToDetailPipeSDSL is available but I cant find the Ultralytics control when doing a search. Have tried removing and re-installing the impact pack but its the same. Appreciate any guidance....

[@sedetweiler]

I would ask the developer. I ran into this as well a few months back and just did a reinstall. I do that every so often as I try a lot of random nodes and I think some leave junk behind.

[@sangboizou2970]

the name is changed to " ONNXDetectorProvider ''

[@DemonPlasma]

forgive me if I'm blind but when i click the link for QR CodeMonster it just takes me to the version for SD2.1, not SDXL.

[@sedetweiler]

There is not SDXL version yet. I keep checking and can't wait for it to pop up.

[@DemonPlasma]

any idea why it takes me 5+ minutes to generate a single SDXL image but I can do SD1.5 images with the same setting in seconds? running on a rtx2080S

[@sedetweiler]

Hmmm, no idea. Is it the same of you use invoke?

[@TransformXRED]

Does it work with the regular controlnet for 1.5 that was released a while back for the qr codes?

[@sedetweiler]

yes, there is one for 1.5 already out and about.

[@TransformXRED]

@@sedetweiler I checked on Civitai, and saw a v2.0 for SD 1.5 (I had the v1 of the controlnet). I'll try your workflow now with that. Thanks :D

[@sedetweiler]

yup! let me know how it goes!

[@TransformXRED]

@@sedetweiler well, the v2 for the 1.5 gives me "meh" results. I tested the V1 (that I used for qr codes in auto1111 months ago), the results are way cleaner. But the results are really cool!

[@jayronnbulworth1915]

I can confirm that the V1, 1.5 version is awesome@@TransformXRED

[@gingercholo]

tis a fun one.

[@henrygrantham5685]

I tried this using control_v1p_sd15_qrcode model on SD 1.5 model and it is generating white messaging instead of black when my input image is black. In other words it is taking the negative of it. This makes it very difficult to read. Anyone know what could be wrong? Thanks.

[@sedetweiler]

Then invert the mask you are passing in. It is a new thing, so don't expect a polished UI without flaws. You are part of the bleeding edge here, and you are helping to shape things.

[@henrygrantham5685]

Thank you for this awesome series. I have learned a ton from it. Inverting the image does work.

[@sedetweiler]

Glad to hear!

[@UncreativeN]

How do i get my Ksampler to show the diffusion?

[@sedetweiler]

in the manager, change the preview to the slow TA version.

[@UncreativeN]

Yeah, I already figured it out. Thanks anyway.

[@LouisGedo]

👋

[@sedetweiler]

Cheers!

[@TheDanEdwards]

"fix the faces" - but what about the hands, the feet, and everything else? These tools are still bad.

[@sedetweiler]

There is a hand one, but it isn't amazing yet. But, I think you will see something soon that will make you happy. ;-)