Just find you channels today, i start yesterday with comfyui, your videos are great and really to the point. I gonna binge watch your channel for the next few days.
That's a great one! Thank you. If you can show how we can use multiple controlNets that would be awesome. They introduced the COntrolNet LoRas. Maybe a ControlNet with LoRa setup tutorial will be more convenient.
ok, the best way ive found is using efficiency nodes and lora/control net stacker, in case you havent found out how yet, results will vary and it may take some tuning to get it right.
a site called civitai.com has lots of that kind of stuff, as well as huggingface.co - generally most control net models are 1.5 and if it says 15 or 1-5 or something like that is compatable with most of big models out there. one that is for SDXL will probably say so in the name.
@@ferniclestix Thank you for your response! It's encouraging to hear that people are successfully using ComfyUi for API applications. If you can't connect me with someone directly, could you perhaps point me toward some resources that explain how to achieve this? I'm very interested in diving deeper into this subject. Thanks in advance!
My advice, go to github for comfyUI or other major custom nodes and ask around. see if anyone is developing an open API or that kind of stuff for it. Can't tell you more than that other than that yes, comfyUI as a backend is a thing. not really able to tell you more. Not a coder my self, Im just really good at workflows and digital art, and teaching people.
Can you do a video for controlnet w/ inpaint module for simple resize + fill using "Pad Image for Outpainting" node? Like say, extending the left and right sides of an image
how do I intall a controlnet model? I downloaded diffusion pytorch model saftensors and put it in the checkpoints folder but when I run, it says undefined
lol youll just have to wait for the advanced tutorial - lol joking. generally you can do all kinds of stuff with the conditioning lines, you could combine both with a combine conditioning node. or you could do them one after the other while keeping the conditioning seperate for each control net.
working with latents is generally... terrible, the information is highly compressed in latent space which makes the outputs kind of uncontrollable. you are better off combining conditioning or images. the tools are better and more developed. there are some custom nodes that do things with latents but its mostly just simple blends and stuff.
it depends how much change to the geometric room you want. I would suggest making use of Compositing if you don't want to mess with the shape of the room but put someone in, ive got a tutorial on that, requires masking and some custom nodes.
@@ferniclestix For example, when we say neoclassical interior design, it naturally gives a neoclassical room. But I don't want that, I want him to give a photo of an empty room and fill it neoclassically. Or I want to apply an interior design that I found on the internet to the empty room I gave, but most importantly, it will not change the geometric shape of the room I gave it.
there is no way to achieve exact results like that without cropping and masking. if you can find the exact style you want and create a prompt to build it then thats great, me, i'd mask out the furniture I wanted to change and apply a second prompt just to that area.
@@haliskurguyan maybe try initial generation in SD 1.5, applying controlnet models MLSD and Depth? And finalize here, using SDXL and refiner? + inpaint if necessary 🤔
Just find you channels today, i start yesterday with comfyui, your videos are great and really to the point. I gonna binge watch your channel for the next few days.
:D happy to help
That's the simplest explanation for controlnet I've seen, well done !
I'm in the early stages of learning about comfyui, it's hard to find short and clear video tutorials like this, thank you very much
Sensational. Quick and to the point, in and out.
Your explanations are so good and clear. Please continue making more content! thank you!!
That's a great one! Thank you.
If you can show how we can use multiple controlNets that would be awesome. They introduced the COntrolNet LoRas. Maybe a ControlNet with LoRa setup tutorial will be more convenient.
be a bit before the longer tutorial of this is out. ill be covering basically everything about control nets in it.
@@ferniclestix sounds great! I'll keep an eye on your channel for the news :)
ok, the best way ive found is using efficiency nodes and lora/control net stacker, in case you havent found out how yet, results will vary and it may take some tuning to get it right.
Thx for explanation
thanks! good info :) One question tho: How come that your KSampler shows the sample steps? :O I was looking for that option for a while.
th-cam.com/video/hdWQhb98M2s/w-d-xo.html here, its snuck away in the basic starter tutorial.
@@ferniclestix thank you :)
Where do you obtain the proper models? I have some files from an extension but I have no idea if they’re compatible with comfyui
a site called civitai.com has lots of that kind of stuff, as well as huggingface.co - generally most control net models are 1.5 and if it says 15 or 1-5 or something like that is compatable with most of big models out there. one that is for SDXL will probably say so in the name.
Once you got a ComfyUi workflow, do you know how to turn it into an API / serverless service so that you can call it from an app?
nope. although i know people who are using it for this kind of thing so its possible.
@@ferniclestix Thank you for your response! It's encouraging to hear that people are successfully using ComfyUi for API applications. If you can't connect me with someone directly, could you perhaps point me toward some resources that explain how to achieve this? I'm very interested in diving deeper into this subject. Thanks in advance!
My advice, go to github for comfyUI or other major custom nodes and ask around. see if anyone is developing an open API or that kind of stuff for it.
Can't tell you more than that other than that yes, comfyUI as a backend is a thing. not really able to tell you more.
Not a coder my self, Im just really good at workflows and digital art, and teaching people.
Can you do a video for controlnet w/ inpaint module for simple resize + fill using "Pad Image for Outpainting" node? Like say, extending the left and right sides of an image
sounds very complex, I'm working on an outpainting one but... yeah, having a strange bug crop up with the node thats hampering things :
how do I intall a controlnet model? I downloaded diffusion pytorch model saftensors and put it in the checkpoints folder but when I run, it says undefined
put them in the control net folder and use a control net loader
@@ferniclestix thank you.
Great video, can you upload your comfy scheme?
Just put in a default workflow and add like 5 nodes, you can just copy off the image. only nodes required I think was WAS node suite.
What about two Controlnet, from two separate images, one for depth, another for canny? 😅
lol youll just have to wait for the advanced tutorial - lol joking. generally you can do all kinds of stuff with the conditioning lines, you could combine both with a combine conditioning node. or you could do them one after the other while keeping the conditioning seperate for each control net.
@@ferniclestix I was rather having trouble with the latents and how/if to combine them. Cheers!
working with latents is generally... terrible, the information is highly compressed in latent space which makes the outputs kind of uncontrollable. you are better off combining conditioning or images. the tools are better and more developed. there are some custom nodes that do things with latents but its mostly just simple blends and stuff.
How can I get image preview in my KSampler like yours?
its a setting if you have comfyUI manager, on the left side you should see some options to choose a preview render method.
There is an empty room, how can I fill it? I am trying to fill an empty room in a certain style without disturbing the geometric shape of the room.
it depends how much change to the geometric room you want. I would suggest making use of Compositing if you don't want to mess with the shape of the room but put someone in, ive got a tutorial on that, requires masking and some custom nodes.
@@ferniclestix For example, when we say neoclassical interior design, it naturally gives a neoclassical room. But I don't want that, I want him to give a photo of an empty room and fill it neoclassically. Or I want to apply an interior design that I found on the internet to the empty room I gave, but most importantly, it will not change the geometric shape of the room I gave it.
there is no way to achieve exact results like that without cropping and masking.
if you can find the exact style you want and create a prompt to build it then thats great, me, i'd mask out the furniture I wanted to change and apply a second prompt just to that area.
@@ferniclestix i am doing it. My goal right now is to fill an empty room the way I want, no matter what I use. Thanks, appreciate it.
@@haliskurguyan maybe try initial generation in SD 1.5, applying controlnet models MLSD and Depth? And finalize here, using SDXL and refiner? + inpaint if necessary 🤔
great stuff