Thank you very much, Mateo! We have tons of nodes, but in general there are not many explanations and workflows with math, loops, conditions, switches. More tutorials and examples on these topics are really helpful to create efficient workflows. Your work is really appreciated!
Finally caught one of your videos early, just wanted to say HUGE THANK YOU for your work, I've improved my workflows so much thank to your nodes and your videos. Thank you!
I have been working with your simple math nodes for a long time, just noticed these new nodes recently, thanks for going through them in an easy to follow way.
Yeah I've often wondered if there was a method to tell comfyUI what to do if a condition is satisfied, and that math node you've shown some videos ago has made me realize that it's actually possible. Thanks for sharing this.
I like your "expression" way of handling math in comfy very much!!! It's a brilliant idea, thank you very much. The original node based math was very clumsy to use. The comparasion might also be an expression, so basically create a ternary node.
Sincerely thank you for sharing all the time. You make every complex technical point so simple and easy to understand, which will benefit me a lot every time.I would like to ask if there is any new breakthrough in ComfyUI's animation production recently. I am looking forward to your sharing some knowledge about animation production in the following. Thank you again! (Computer translation, may not be accurate)
Hi Matteo, I hope you are doing well. I have been doing some experiments with the flux model and I would like to share some observations with you. I would appreciate it if you could comment, confirm or deny these: 1. Flux really does a good job at high resolutions, without any artefacts like double characters as we have with SD 1.5, 2 etc. 2. The generation speed for higher resolution images is not linear, that means, that a 1920x1080 generation is much faster than 4 960x540 generations. 3. Higher resolutions really improve the details on the generation. My suggestion would be to go as high of a resolution as possible rather than using a tiled upscaler. I am curious to hear your experience and thoughts on this.
1. at higher resolution Flux doesn't hallucinate. they say it's good up to 2048x2048 2. mh, I can't say I noticed that, there are resolutions flux is better at for sure 3. at higher resolution you don't get hallucination but not many more details. noise injection helps
cosa dire dei tuoi tutorial! forse solo un grazie non basta..... gli ho attaccato un LoRA con un training su un prodotto e cazzarola sia a un punto di svolta definitivo. GRAZIE
Hi Matteo, thanks for all the work put into these videos, they're invaluable! Looking at the way the upscale was split, was there any particular reason for this? Testing the workflow I couldn't see the benefit of running an upscale over two separate parts versus the image as a whole. Am I missing something here? Thanks!
Gemini said you covered it but I'm not really sure that you did.. because artist. Math. And if I saw these two magic numbers in your transcript, maybe it was covered: 512 x 512: 262,144 pixels * 1024 x 1024: 1,048,576 pixels. But I didn't. So....: Lol, it's like I listen to the entire video intently. And then when you're done, I go: " what?" 🤯 Anyway, what I was wondering if you covered this and if not maybe a great video idea, especially with people like myself with 8 vram are always keeping that pixel count within bounds so you don't realize after 5min your command window just closed😢, And there's that little reconnects box of death in the corner of comfy (dear comfy, UI interface with a huge crash dialog box and claxon alert when this happens?😂). I mean I created a notepad with the usual picture frame sizes with chat GTP to give me the correct pixel accounts for the equivalent for 512x512 and 1024x1024 at different ratios. (It's a handy list for a 3x4, 16x9 etc. But if you do do that double check cuz Chatgtp will make it look like it knew what it was doing but I had to ask it to redo the math a couple times) . Anyway, I'm just wondering if you can utilize nodes like this....
Interesting looking nodes, BTW is there any way to package nodes into "boxes" with inputs and outputs like in some other software to avoid copy pasting several nodes, as Comfy graphs seem awfully messy.
@@latentvision Not talking about groups, talking about something like group, but which you can store as a prefab, with required inputs and outputs. So one doesn't have to build or copypaste same kind of setups repeatedly.
For example, if you know Amplify Shader editor for Unity game engine, it had something similar, quoting their manual (as it illustrates the concept quite well) - "Shader Functions are a great way to reduce canvas clutter by packing complex networks into single nodes, and a great way to eliminate unnecessary repetitive tasks; they can be considered a form of "grouping", that offers you a way to group nodes into more manageable asset". So it is not important that I used shaders as example, it could be anything node based, but I think this concept would translate pretty much as is for Comfy.
great work , as always , i want to share you some of my result , sd 1.5 is not dead , i think it can compite with ai generartor out there , so u might help me , so how i can send to u something private
Good video, but didn't find it helpfull for myself. Prefer to use more high level nodes, or more advanced nodes like code node if i am actually need to go full math mad
Best AI Channel on RUclips. Thanks for the video!
Thank you very much, Mateo! We have tons of nodes, but in general there are not many explanations and workflows with math, loops, conditions, switches. More tutorials and examples on these topics are really helpful to create efficient workflows. Your work is really appreciated!
This is truly the most informative Stable Diffusion channel on RUclips. Thank you for your work.
This is the channel to go for anybody who wants to dig deeper into the way comfyUI works while understanding what's behind the process. Fantastico!
Finally caught one of your videos early, just wanted to say HUGE THANK YOU for your work, I've improved my workflows so much thank to your nodes and your videos. Thank you!
Very well explained and particularly useful for utilizing Comfy's full potential. Thank you for the lesson. :)
I have been working with your simple math nodes for a long time, just noticed these new nodes recently, thanks for going through them in an easy to follow way.
Commento tattico per l'algoritmo e per dirti grazie Matteo!!!
ahaha grazie :D
Can't wait for loops!
This is too much spaghetti even for an Italian. 😂
🍝🍝🍝
Yeah I've often wondered if there was a method to tell comfyUI what to do if a condition is satisfied, and that math node you've shown some videos ago has made me realize that it's actually possible. Thanks for sharing this.
Your instructions are really detailed and effective. I always look forward to your new videos; they are so helpful
I like your "expression" way of handling math in comfy very much!!! It's a brilliant idea, thank you very much. The original node based math was very clumsy to use. The comparasion might also be an expression, so basically create a ternary node.
pro tip: you can use a Math Node as a comparison formula. a>0 a>b a!=b... they all work. As long as it returns 0 or 1 you can use whatever
"too much spaghetti, even for an Italian."
🤌
I was looking for this comment 😂 there are never too many spaghetti for us 😂
Mateo's videos are on another level, really awesome!
Thanks for the video Matteo!
This is AWESOME Mateo - way to go!
Sincerely thank you for sharing all the time. You make every complex technical point so simple and easy to understand, which will benefit me a lot every time.I would like to ask if there is any new breakthrough in ComfyUI's animation production recently. I am looking forward to your sharing some knowledge about animation production in the following. Thank you again! (Computer translation, may not be accurate)
we are all waiting for Flux Video... that appears to be very interesting. to be determined if we will be able to use it locally
Thank you!
Thank you so much for again another great video!! 😊
love this, thanks man
7:10 That escalated wuickly 😵
Hi Matteo,
I hope you are doing well. I have been doing some experiments with the flux model and I would like to share some observations with you. I would appreciate it if you could comment, confirm or deny these:
1. Flux really does a good job at high resolutions, without any artefacts like double characters as we have with SD 1.5, 2 etc.
2. The generation speed for higher resolution images is not linear, that means, that a 1920x1080 generation is much faster than 4 960x540 generations.
3. Higher resolutions really improve the details on the generation. My suggestion would be to go as high of a resolution as possible rather than using a tiled upscaler.
I am curious to hear your experience and thoughts on this.
1. at higher resolution Flux doesn't hallucinate. they say it's good up to 2048x2048
2. mh, I can't say I noticed that, there are resolutions flux is better at for sure
3. at higher resolution you don't get hallucination but not many more details. noise injection helps
che bello cercare ai e trovare un italiano che lo spiega
Great!
Merci !
Been waiting for this!
cosa dire dei tuoi tutorial! forse solo un grazie non basta..... gli ho attaccato un LoRA con un training su un prodotto e cazzarola sia a un punto di svolta definitivo. GRAZIE
it return 512 in portrait mode :) because in PM width must be lower then height.
Hi Matteo, thanks for all the work put into these videos, they're invaluable! Looking at the way the upscale was split, was there any particular reason for this? Testing the workflow I couldn't see the benefit of running an upscale over two separate parts versus the image as a whole. Am I missing something here? Thanks!
hey thanks. generally tiling allows higher upscale ratios
Gemini said you covered it but I'm not really sure that you did.. because artist. Math. And if I saw these two magic numbers in your transcript, maybe it was covered: 512 x 512: 262,144 pixels
* 1024 x 1024: 1,048,576 pixels.
But I didn't.
So....:
Lol, it's like I listen to the entire video intently. And then when you're done, I go: " what?" 🤯
Anyway, what I was wondering if you covered this and if not maybe a great video idea, especially with people like myself with 8 vram are always keeping that pixel count within bounds so you don't realize after 5min your command window just closed😢, And there's that little reconnects box of death in the corner of comfy (dear comfy, UI interface with a huge crash dialog box and claxon alert when this happens?😂).
I mean I created a notepad with the usual picture frame sizes with chat GTP to give me the correct pixel accounts for the equivalent for 512x512 and 1024x1024 at different ratios. (It's a handy list for a 3x4, 16x9 etc. But if you do do that double check cuz Chatgtp will make it look like it knew what it was doing but I had to ask it to redo the math a couple times) . Anyway, I'm just wondering if you can utilize nodes like this....
I am grateful..
You are awesome. thanks man.
no, you are awesome!
Have you seen the "Griptape" nodes ?
Maybe they are useful ?
Top 👍
Hello Mateo this is very interesting. What is your suggestion to incorporate LORA in your workflow ?
I know people are already using it with loras so it's just a matter of connecting it to the model pipeline
Yeeeaaaahhhh!!!! ❤ so cool
Interesting looking nodes, BTW is there any way to package nodes into "boxes" with inputs and outputs like in some other software to avoid copy pasting several nodes, as Comfy graphs seem awfully messy.
yes you can group nodes from the right click menu
@@latentvision Not talking about groups, talking about something like group, but which you can store as a prefab, with required inputs and outputs. So one doesn't have to build or copypaste same kind of setups repeatedly.
For example, if you know Amplify Shader editor for Unity game engine, it had something similar, quoting their manual (as it illustrates the concept quite well) - "Shader Functions are a great way to reduce canvas clutter by packing complex networks into single nodes, and a great way to eliminate unnecessary repetitive tasks; they can be considered a form of "grouping", that offers you a way to group nodes into more manageable asset". So it is not important that I used shaders as example, it could be anything node based, but I think this concept would translate pretty much as is for Comfy.
@@devnull_ yes, it's called "convert to group nodes"
King shit. Thank you.
great work , as always , i want to share you some of my result , sd 1.5 is not dead , i think it can compite with ai generartor out there , so u might help me , so how i can send to u something private
How do you get the fancy sliders ? 🤔
it's a node in the comfyui essentials
I don't think in the comments.
Wait did he just say LOOPS in ComfyUI???!! Wasn't that impossible?
it is possible now!
@@latentvision 🤯
B.R(a-v)0
Good video, but didn't find it helpfull for myself. Prefer to use more high level nodes, or more advanced nodes like code node if i am actually need to go full math mad
why is maths everywhere...
because it's kwel!