Ok, so we have UA-camrs who teach AI generations, and then we have Matteo, who really understand and has a deep knowledge of what he's using. You play in a different Championship. Thank you
Here is the master brain of ComfyUI , everything seems so evident to you, on my part I just need to review 3 or 4 times your video , stop every 15 sec , take a note as a reminder and spend 2h at least, to really assimilate all the knowledge given here. It is such a gift you are sharing here with us. Thanks again for your generosity and sharing knowledge
Thank you so much, Matteo, you really taught me to get to the core of SD gradually. I just watched this video for the first time and there is still a lot to digest and I will have to watch it a few more times to really get a handle on the workflow.
I watch your videos many times and your teachings are a great pleasure to me. And I also feel my skills improving. Thank you so much for your teachings
thank you @latentvision for the inspiration. Yes , I am totally agree about the Lineart. I will try it out to mask the background like how you mentioned. 👍
Thank you. The method worked for me after several days of trying by following your explanation and building the worklow instead of downloading it to understand each part accurately
Your knowledge and approach and how you think are most inspirational to me, I strive to be more like you if only in a small way. These tutorials are like good food!
If you want to do more videos about animation in ComfyUI I would be happy to watch! Always such a lot of knowledge in your videos, i love it! You come here looking for one thing and learn three others on the way!
Hey Mateo, thanks for the great video, I was able to learn a lot again especially in terms of optimization. The max resolution of ClipVision was also new to me. 👍👍👍
Very interesting, thanks. Is there a way in comfyUI to make a 3d object integration in a video matching both the video lighting and the video style? No one seems to be capable of until now.
Just out of curiosity, you cropped the 488 image into two pieces at 224 rather than 244 which would have been half the original. Was there a reason in particular or just a "close enough" sort of thing? Still learning here, but loving it. :) Thanks!
no it's not an error, I downscaled it to 488 but I'm taking only 224 so the ballerina is actually a pinch bigger (ie: I'm cutting out the sides of the animation and concentrating only on the main character). Depending on the video you can totally downscale to 448 that will give you the whole frame. glad you noticed it though
another useful technique even explained in one sentence, thanks again matteo. you are developping and teaching non stop, you can not be a human! @@latentvision
Thank you, Matteo!!!! You are so amazing. I am every time feeling excited in advance when i see a new video on your channel. How do you think it might be useful to create a node that would accept an image of any proportions and prepare it for the clip vision internally, splitting it to squares, and one ip-adapter node that could accept its output?
hey thanks! Yes, I'm thinking of adding an "auto-tile" node, but it could be expensive computationally if you add a lot of tiles. I have to think about it but it's doable.
Haha you are and honestly thank you for sharing your knowledge. I’m new to this. I’m a developer by trade by delving into this world seemed overwhelming until I came across your videos. So thank you again :)
12:19 hey Matteo, what if I want to keep only blue part instead of red from Mask from Color. how can I set it up? I changed around numbers and I kept getting the entire piece black. thank you so much for the amazing tutorial!
load the mask into a paint program and with the eyedropper select the color you want to replace. check the RGB values and you are done. increase the threshold by one or two just to be safe
I resize the image slightly bigger than I need and then crop out the 224x224 tile. That way I get slightly more details on the parts I'm interested into and also crop out some of the background on the sides. I'm sorry I didn't explain that in the video. But yes, generally you want to crop it at 448
This is wonderful! Thank you for showing your techniques. May I ask how to use controlnet tile on a specific region? I used the segmented mask technique you showed, but since the empty space is black, controlnet tile makes that region black too. How to tile control a specific region without making the rest of the image black? I'd really appreciate any tips.
THANK YOU THANK YOU THANK YOU I have not tested the part where you do every part of the video and put it into 16frames sections, but I will be re-watching that after I make up breakfast I was wondering why things changed when I changed the amount of frames. You make for an amazing teacher I watched the "Image stability and repeatability (ComfyUI + IPAdapter)" and "Animations with IPAdapter and ComfyUI" I didn't learn much on those because I wasn't using comfyui at the time and I fell asleep on the FaceID video because I was tired not because of the content so I don't remember even watching it and then this Video now that I can do something interesting animated things in comfy then I watch this one and it solved my biggest issue(still crossing my fingers but I think so because of how you explained it) and also goes slow enough to help teach how to do some node play/work. I'd like to see more and more content as things evolve which they are rapidly I hope you can keep everyone up to date because I think you can do it better than most.
Okay so using the Uniform Context Options and mixing that up gets me interesting results my pc is a beast so it isn't too bad but damn I can't wait till things evolve. I'm going to be able to make a movie soon using this technology. I can't believe I can actually do this stuff. Anyways I'm stuck and want certain results I'll be back here later maybe tomorrow or tonight with some questions crossing my fingers that sometime during the holiday weekend you have free time. Happy New year I hope I can ask you for help also I subscribed to you not like that is a good trade-off for some extra help maybe I could buy you a beer or something. I could also get you some drone footage in the summer/winter or something I don't know maybe you will have an idea.
Amazing result and super clear explanation on this tutorial, thank you Mateo! One question, I didn't understand "Uniform context options" node very well, which looks like a parameter on AnimateDiff loader, what is it for?
models are trained at 16 frames usually (now we have longer models though) the "context options" renders longer videos by computing 16 frames at the time.
Hey, how can I say combine this with more adapters, if I have two things I want to transition between I use a transition mask, but how do I also split the image and give each part more continuity, it would be like 4 ipadapters in total. I'm also using two Face Id masks that apply to specific faces. Any ideas to manage all these attention masks?
I'm sorry this is not something I can explain in a YT comment. Check my ipadapter animations video for a simple transition workflow. Then yes you can add more areas with multiple masks... but maybe using other techniques would be better (like tile controlnet)
Haha thanks for the reply anyway. I figured it out. It ended up involving a lot of mask manipulation. I used two tile controlnets to keep the start and end consistent and multiple ip adapters and masking to interpolate between concepts. The masks get more complex as I add more IPAdapters, and it does take a lot of fine tuning.@@latentvision
Great. In the new IP Adapter advanced is no "unfold batch" option anymore. is this obselete with V2 or do you need to use the batch version of the IP adapter instead? Thank you.
i was trying to follow your tutorial step by step instead of copying the workflow for learning purpose but at some point I got this and I can't move any forward: Error(s) in loading state_dict for Resampler: size mismatch for proj_in.weight: copying a param with shape torch.Size([768, 1280]) from checkpoint, the shape in current model is torch.Size([768, 1024]).
When I have a nice setup (images) and implement Animdiff (TEXT2VIDEO) it always changes and get very blurry. How could I use IPAdapter to get my text2video sharpen again?
Hi I have some issues with the AdnimateDiffLoader. It gives me the error like below Error occurred when executing ADE_AnimateDiffLoaderWithContext: module 'comfy.ops' has no attribute 'Linear' Ive used my own workflow and yours but it seems to have the same issues. Could you let me know if there is a workaround?
Hi, I get this error "When loading the graph, the following node types where not found: ImageCrop+ ImageCASharpening+ MaskFromColor+ MaskBlur+ ImageResize+ Nodes that have failed to load will show as red on the graph." Although I installed them via Install Missing CustomNodes and restarted them. But there is no result
You are the best! thank you very much. I'm getting error maybe you can please help? RuntimeError: mat1 and mat2 shapes cannot be multiplied (1232x768 and 1024x320)
Can I use this workflow with macOS? Macbook Pro M1 14º I'm getting this error Error occurred when executing KSampler: The operator 'aten::upsample_bicubic2d.out' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature On my windows PC I have only a 1050ti GPU and it's not enough to handle this workflow.
By the way ... I would like to endorse most positive comments of people about your channel. From all the channels I've seen yours is by far the most innovative and easy to follow. Maybe it is because it seems that your really know the tech and is more comfortable to pass a way your knowledge to us.
@latentvision : I want to replicate your workflow as a starting point with this, but the ballerina video is unfortionately not available atm. Would it be too presumptuous to ask for upload that file somewhere?
@latentvision ok, for me still not, no matter which browser I use, still after click on that download button just : "Video temporarily unavailable" ... thx for your interest
Man you are out of this world. Really i do not have words to describe what you have unlocked for this world. Thank you. can you please share your discord or any chat group where i can join your community.
I am trying comfyUi for days now, mybe not skilled enough, but thing is not there yet. most nodes and models needed cant be found anywhere. I hope this will one day work better for semi pro users
I did a 6-piece landscape for a ratio of 2:3. (640x480 input with six 224x224 pieces to cover 672x448). I'm not sure why I felt the need to post this? I guess it's the excitement of learning comfyui.
I try to use your workflow after watching your jellyfish ballerina animation with animatediff video. But "comfyui's controlNet auxiliary preprocessors and comfyui-videoHelperSuite" node fall to load (import failed). I am installing via comfy manager also try to install manually but problem was to solved. I am using Amazon sagemaker studio lab. Please help me to fix this issue.
Ok, so we have UA-camrs who teach AI generations, and then we have Matteo, who really understand and has a deep knowledge of what he's using. You play in a different Championship. Thank you
no, we have youtubers exploring and reviewing ai
Here is the master brain of ComfyUI , everything seems so evident to you, on my part I just need to review 3 or 4 times your video , stop every 15 sec , take a note as a reminder and spend 2h at least, to really assimilate all the knowledge given here. It is such a gift you are sharing here with us. Thanks again for your generosity and sharing knowledge
This guy's skills are approaching godlike.
I'm not worthy 😅
@@latentvision❤
so it's why he's the developer
Thank you so much, Matteo, you really taught me to get to the core of SD gradually. I just watched this video for the first time and there is still a lot to digest and I will have to watch it a few more times to really get a handle on the workflow.
I watch your videos many times and your teachings are a great pleasure to me.
And I also feel my skills improving.
Thank you so much for your teachings
Ah! Mateo, I don't know if it's your accent, your knowledge, the way you transmit it, your tranquility ... But I'm falling in love baby! ❤❤
Me tooooooo ❤
I've never enjoyed watching tutorials as much as I have yours. Like, for anything. You are a great teacher.
Brilliant job, Matteo, and as always, beautifully explained. I love your work. Thanks for sharing it with us.
thank you @latentvision for the inspiration. Yes , I am totally agree about the Lineart. I will try it out to mask the background like how you mentioned. 👍
Thank you. The method worked for me after several days of trying by following your explanation and building the worklow instead of downloading it to understand each part accurately
Your knowledge and approach and how you think are most inspirational to me, I strive to be more like you if only in a small way. These tutorials are like good food!
Your videos have so much valuable information. You're a master at what you do and you deserve way more views. Thank you for your work!
Fantastic! Thank you for sharing the process in such great detail and workflow.
It's beautiful. I hope there will be more animated works, such as animated picture stories
Maestro Latente!!! Absolutely masterful!! Bravoooooo!!! ❤️🇲🇽❤️
You are amazing!!!! I’m in complete awe of your skills and knowledge
My first impressions of comfy ui was horrible. But something in me told that this tool will be powerful and feature rich. You've proven this 🎉
This was in March of 2023, now in the end 😮
Awesome. You are sensational with ComfyUI
Matteo, thank you for this very impressive and inspiring tutorial! Keep on your extraordinary work.
If you want to do more videos about animation in ComfyUI I would be happy to watch! Always such a lot of knowledge in your videos, i love it! You come here looking for one thing and learn three others on the way!
This channel is so good. Got a totally different perspective on how to use comfyUI. Well done!
Very Inspiring - Thank You
Your technique is flawless victory. Well done!
Hey Mateo, thanks for the great video, I was able to learn a lot again especially in terms of optimization. The max resolution of ClipVision was also new to me. 👍👍👍
This is just amazing Mateo! 🤯
Many thanks for your sharing and teaching, you really understand all the nodes in ComfyUI..Bravo Matteo!
Very interesting, thanks.
Is there a way in comfyUI to make a 3d object integration in a video matching both the video lighting and the video style?
No one seems to be capable of until now.
Just out of curiosity, you cropped the 488 image into two pieces at 224 rather than 244 which would have been half the original. Was there a reason in particular or just a "close enough" sort of thing? Still learning here, but loving it. :) Thanks!
no it's not an error, I downscaled it to 488 but I'm taking only 224 so the ballerina is actually a pinch bigger (ie: I'm cutting out the sides of the animation and concentrating only on the main character).
Depending on the video you can totally downscale to 448 that will give you the whole frame.
glad you noticed it though
Thanks for the explanation @@latentvision :)
another useful technique even explained in one sentence, thanks again matteo.
you are developping and teaching non stop, you can not be a human!
@@latentvision
Absolutely amazing tutorial. Thanks for sharing
can't seem to find DW preprocessor like the one you have in this video just the basic version where it only has hand, body and face.
I think we can use xmem or cutie to create masks easily instead of cocc segmenter, etc (mentioned in 11:37)
Grande Matteo! Grazie per il tuo lavoro, per la passione e la condivisione.
Thank you, Matteo!!!! You are so amazing. I am every time feeling excited in advance when i see a new video on your channel. How do you think it might be useful to create a node that would accept an image of any proportions and prepare it for the clip vision internally, splitting it to squares, and one ip-adapter node that could accept its output?
hey thanks! Yes, I'm thinking of adding an "auto-tile" node, but it could be expensive computationally if you add a lot of tiles. I have to think about it but it's doable.
thank you so much for all of your work!
honestly you are amazing! incredibly smart! thank you for your videos!
nnnaaah I'm not!
Haha you are and honestly thank you for sharing your knowledge. I’m new to this. I’m a developer by trade by delving into this world seemed overwhelming until I came across your videos. So thank you again :)
12:19 hey Matteo, what if I want to keep only blue part instead of red from Mask from Color. how can I set it up? I changed around numbers and I kept getting the entire piece black. thank you so much for the amazing tutorial!
load the mask into a paint program and with the eyedropper select the color you want to replace. check the RGB values and you are done. increase the threshold by one or two just to be safe
@@latentvision thank you!!!
SPETTACOLARE
A great mentor
You are my source of motivation, and I will work hard to learn.
It's incredibly great👍
Brilliant again matt3o, thank you. query at 7:50 you put size 488x488, did you mean 448 as in 2x224?
I resize the image slightly bigger than I need and then crop out the 224x224 tile. That way I get slightly more details on the parts I'm interested into and also crop out some of the background on the sides. I'm sorry I didn't explain that in the video. But yes, generally you want to crop it at 448
thank you for the explanation@@latentvision
i love this workflow. hopefully, you can discuss in improving the face detail
yes, increasing details and sharpness is next thing we need to cover... so many things to do 😄
wow youre the best ai guy for comfyui
do you have plans in creating workflow for old photo restoration?
yes, that is a very interesting topic!
This is wonderful! Thank you for showing your techniques. May I ask how to use controlnet tile on a specific region? I used the segmented mask technique you showed, but since the empty space is black, controlnet tile makes that region black too. How to tile control a specific region without making the rest of the image black? I'd really appreciate any tips.
Thanks! ACN Advanced Control Net has a "mask" option
You are Such a great value to the community 😊 Many thanks 🙏
Great tutorial Matteo!
I WICH YOUR VIDEOS NEVER END
maybe its time to try animation out, the greatest of all time made a tutorial ❤❤
nice work, 👍👍👍
Thank you!
THANK YOU THANK YOU THANK YOU I have not tested the part where you do every part of the video and put it into 16frames sections, but I will be re-watching that after I make up breakfast I was wondering why things changed when I changed the amount of frames. You make for an amazing teacher I watched the "Image stability and repeatability (ComfyUI + IPAdapter)" and "Animations with IPAdapter and ComfyUI" I didn't learn much on those because I wasn't using comfyui at the time and I fell asleep on the FaceID video because I was tired not because of the content so I don't remember even watching it and then this Video now that I can do something interesting animated things in comfy then I watch this one and it solved my biggest issue(still crossing my fingers but I think so because of how you explained it) and also goes slow enough to help teach how to do some node play/work. I'd like to see more and more content as things evolve which they are rapidly I hope you can keep everyone up to date because I think you can do it better than most.
Okay so using the Uniform Context Options and mixing that up gets me interesting results my pc is a beast so it isn't too bad but damn I can't wait till things evolve. I'm going to be able to make a movie soon using this technology. I can't believe I can actually do this stuff. Anyways I'm stuck and want certain results I'll be back here later maybe tomorrow or tonight with some questions crossing my fingers that sometime during the holiday weekend you have free time. Happy New year I hope I can ask you for help also I subscribed to you not like that is a good trade-off for some extra help maybe I could buy you a beer or something. I could also get you some drone footage in the summer/winter or something I don't know maybe you will have an idea.
H ! amazing workflow - so much better than anything else out there! Do you mind telling us where do download the ip adapter image encoder sd15?
please check the extension repository on github
Incredible Work !! Thnqks so much
i'm wondering how u know the parametres so deeply, can u create avideo for tuning the parametres to get the desired results?
Many thanks
Amazing result and super clear explanation on this tutorial, thank you Mateo! One question, I didn't understand "Uniform context options" node very well, which looks like a parameter on AnimateDiff loader, what is it for?
models are trained at 16 frames usually (now we have longer models though) the "context options" renders longer videos by computing 16 frames at the time.
Thanks for explanation!@@latentvision
Hey, how can I say combine this with more adapters, if I have two things I want to transition between I use a transition mask, but how do I also split the image and give each part more continuity, it would be like 4 ipadapters in total. I'm also using two Face Id masks that apply to specific faces. Any ideas to manage all these attention masks?
I'm sorry this is not something I can explain in a YT comment. Check my ipadapter animations video for a simple transition workflow. Then yes you can add more areas with multiple masks... but maybe using other techniques would be better (like tile controlnet)
Haha thanks for the reply anyway. I figured it out. It ended up involving a lot of mask manipulation. I used two tile controlnets to keep the start and end consistent and multiple ip adapters and masking to interpolate between concepts. The masks get more complex as I add more IPAdapters, and it does take a lot of fine tuning.@@latentvision
I just got on your video, and I'm not sure why i watched other videos before haha 😂
Crazy good video sir 👌👌👌
Great. In the new IP Adapter advanced is no "unfold batch" option anymore. is this obselete with V2 or do you need to use the batch version of the IP adapter instead? Thank you.
the batch nodes are for animations yes
i was trying to follow your tutorial step by step instead of copying the workflow for learning purpose but at some point I got this and I can't move any forward:
Error(s) in loading state_dict for Resampler:
size mismatch for proj_in.weight: copying a param with shape torch.Size([768, 1280]) from checkpoint, the shape in current model is torch.Size([768, 1024]).
check IPAdapter repository for help github.com/cubiq/ComfyUI_IPAdapter_plus
When I have a nice setup (images) and implement Animdiff (TEXT2VIDEO) it always changes and get very blurry. How could I use IPAdapter to get my text2video sharpen again?
Hi I have some issues with the AdnimateDiffLoader. It gives me the error like below
Error occurred when executing ADE_AnimateDiffLoaderWithContext:
module 'comfy.ops' has no attribute 'Linear'
Ive used my own workflow and yours but it seems to have the same issues. Could you let me know if there is a workaround?
I'm sorry it's hard to say, I'd suggest to upgrade comfyui and the extensions
Error occurred when executing OneFormer-COCO-SemSegPreprocessor:
No module named 'controlnet_aux.oneformer' how can I fix that :( ?
Das war wirklich wunderschön und ausnehmend informativ. Vielen, vielen Dank 👍😍
any reason why you use the 1.5 v2 motion model for animatediff over v3?
well v3 wasn't out when I started making this video 😄You are free to try it anyway, it should work.
@@latentvision I'm really happy with the results. Thanks for your hard work compiling this information :)
where i can download 3 ?
Hi, I get this error
"When loading the graph, the following node types where not found:
ImageCrop+
ImageCASharpening+
MaskFromColor+
MaskBlur+
ImageResize+
Nodes that have failed to load will show as red on the graph."
Although I installed them via Install Missing CustomNodes and restarted them. But there is no result
you probably need to update comfyui
@@latentvision I'll try to update. But it seems in my Colab ComfyUI is updated automatically after each launch
that's so good thanks bro 🙌
You are the best! thank you very much. I'm getting error maybe you can please help? RuntimeError: mat1 and mat2 shapes cannot be multiplied (1232x768 and 1024x320)
mmh you are probably using the wrong combination of models (like the wrong clip vision or the wrong checkpoint)
Amazing. YOu are very intelligent human. :)
I am overwhelmed with so much information, wish i can borrow your brain .. i cant follow.. i have to watch multiple times...
Can I use this workflow with macOS? Macbook Pro M1 14º
I'm getting this error
Error occurred when executing KSampler:
The operator 'aten::upsample_bicubic2d.out' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature
On my windows PC I have only a 1050ti GPU and it's not enough to handle this workflow.
I need the full error message please. I believe I know what the problem is but I need to be sure
By the way ... I would like to endorse most positive comments of people about your channel. From all the channels I've seen yours is by far the most innovative and easy to follow. Maybe it is because it seems that your really know the tech and is more comfortable to pass a way your knowledge to us.
Outstanding Matteo 🔥
@latentvision : I want to replicate your workflow as a starting point with this, but the ballerina video is unfortionately not available atm. Would it be too presumptuous to ask for upload that file somewhere?
seems to be working www.pexels.com/video/person-woman-girl-steps-4990427/
@latentvision ok, for me still not, no matter which browser I use, still after click on that download button just : "Video temporarily unavailable" ... thx for your interest
yep and one day later > now the download working.... for anyone still interested...
you are awesome :)
nah, you are awesome!
I like your videos the most.❤
Man you are out of this world. Really i do not have words to describe what you have unlocked for this world. Thank you. can you please share your discord or any chat group where i can join your community.
thanks! my discord: latent.vision/discord
I am trying comfyUi for days now, mybe not skilled enough, but thing is not there yet. most nodes and models needed cant be found anywhere. I hope this will one day work better for semi pro users
Is this possible with a 16:9 video or will I run into problems with CLIP Vision?
this video was portrait mode, but it works the same for landscape (16:9 included)
I did a 6-piece landscape for a ratio of 2:3. (640x480 input with six 224x224 pieces to cover 672x448). I'm not sure why I felt the need to post this? I guess it's the excitement of learning comfyui.
Wow this is incredible
magic! :)
+1 subscriber, thank you
Amazing. Thank you
great! as always
Nice work
Gracie 😊
Why can't you loop it?
of course you can! :)
Just wonderful 😍💪🏻✌🏻
I try to use your workflow after watching your jellyfish ballerina animation with animatediff video. But "comfyui's controlNet auxiliary preprocessors and comfyui-videoHelperSuite" node fall to load (import failed). I am installing via comfy manager also try to install manually but problem was to solved. I am using Amazon sagemaker studio lab. Please help me to fix this issue.
bro, youy nuked it, gg
👍👍👍👌👌👌👌❤❤
Oh hey! Still here?
I think so
I quit
Thanks, great job
there was only one problem with loading the video
approached 768x1664