I was also existed about the comfy repo getting updated but the results seem worse than before. I tested the official repo locally with the gradient UI which has also implemented video input and the results are a lot better. I know they still rely on insight face but I would love to hear your opinion about the differences
I tired following your guide and watched all the beginner videos you linked (since I started from zero) but how do I get my webcam to work with the live webcam example? There is no node to choose my webcam source or something.
hi, basically, i had used ur workflow of liveportrait for webcam, but there is an error , Cannot execute because a node is missing the class_type property.: Node ID '#205', so, may u know what custom nodes were missing?
The problem with all these local versions, is their installers and interfaces are usually... abysmal. Like the programs just got it running, and then released it all unpolished and with a long drawn out nasty install process full of dependencies and other library cruftware you have to install along the way just to get it running.
The pros of the local versions is they are more often than not, not stupldly censored, and what they can do today for free for youthey should do for a long time to come. A lot of the online AI tools are free to use not in a limited capacity while they are rightly so in beta (because the results you get with AI is more often miss than hit)... but when they do go pay, you realize, they want yearly subscriptions, and it's in no way worth it... to use the same beta crppy tool you were using before, that did not improve, they just finally finished the subscription billing code. So I'm all for pushing for more open source AI stuff, but the developers really need to focus on polishing up their installers.
@@choppergirl I agree with all of that. Also the 'credits' model I don't like. As you said, most results are miss not hit. So you burn through credits trying to get something useful.
Nerdy-sensei, a good post-processing tool would be facefusion since it can lip sync better, use GPEN for face improvement and also has frame enhancer. Love your work
Where's the Face-Alignment workflow? The examples folder only has workflows with MediaPipe and InsightFace, and there's no Face-Alignment node in the manager.
@@AB-wf8ek Anyone could claim "papers" without actually having anything real. I don't know if that actually happens, but I do know that I've seen soo many "vapor-ware" announcements on so-called amazing "papers" that I unsubscribed and blocked "2 minute papers" because it was a complete waste of my time. "hey look at this cool thing...... ... that nobody will ever get to use" over and over and over... It's getting excited for empty promises. I could CGI render something and make outrageous claims about instant/realtime stuff and claim "2 more papers down the line, you will be really shocked" and just leave it at that. Fool me once...
Comfy: tool for people with brains wired like spaghetti. As if those 'nodes' were anything else than simple function calls with parameters that would be way more readable - and much better to debug - in code form instead of this mess of a GUI. And in fact 'install missing custom nodes' too often fails to work. What non-programmers don't realize is, that moving nodes around like in a kids games has no advantages compared to having code with definite positions from top to bottom. The later being way more transparent and more easy to follow. And easily searchable and replaceable at that.
@@joechip4822 Sadly Comfy and Stability AI are worried that the regular normal person in the street will be able to make content that rivals bug studios, they needed to put a stop to this. That's why thy have been pushing Comfy UI which is only aimed at a minority of people with understanding of nodes. This keeps takes it away from the regular people who helped build and test the ai and back into the hands of the big studios.
@joechip4822 I'm not a programmer, but I am a digital artist. Node based interfaces have been around for decades. Modular rack synth systems are the original manifestations. You can also find node based editing in all the most advanced software for 3D animation, video compositing, video effects, and music. Being able to visually patch input/output connections is profoundly more efficient for editing non-linear workflows and complex custom processes. It's perfectly valid to say you don't like it, but to claim it's not a good system altogether simply disregards everyone who uses it. It's not like there aren't other options. Fooocus and Invoke seem like great alternatives.
@@AB-wf8ek okok, but do you know what had been around also for quite some time? Wire routing for electronic chips and PCBs - which by principle avoids overlapping wires/cables. If comfy had such an option it would reduce the messy 'everybody does it their own way', 'everything is over the place' and 'I need the biggest screen ever to grok this' approach of comfy. A GUI like this simply needs much more tool functions. And by the way... What makes me suspicious is, that newbies and seasoned AI artists alike very often despise such node based GUI - reddit has many debates about this, and you can even find the opinion that these systems are the equivalent of LINUX in their ways of making easy things complicated while mainly allowing the da boys to boast about their own ingenuity. The n.b.d. may come from a time when things were totally analog and machines had actual plugs. But working with a famous modern software synt named 'Pigments' I know that on a digital screen there are better ways, if you don't want to work with code, and the many tools that exist to make this effective. But granted... It may be mainly a matter of taste and experience in a certain field. Especially artists are very often extremely original in their art products, but cling tightly to a certain limited tool set, even if it hadn't been optimized for quite a long time (Photoshop, Ableton Live etc.)
open source forever!
WOW. Now we are talking! Thanks for videos/tuts. Top notch as always... :) days.
Glad you like them!
How to avoid the continuous face size change that completely kills the realism of the reenactment?
How can you stop the head from getting bigger and smaller constantly?
That head wobbling though, kills the effect.
Its not an issue with the gradio app version. Not sure whats up with his results. I assume something in the cozyui spaghetti
nice video. Do you know how to get rid of the flickering? because the head isnt standing correct, it looks bad
The gradio app version seems less jittery.. But I guess using this in cozy is easier for some.
Awesome video mate ❤
I was also existed about the comfy repo getting updated but the results seem worse than before. I tested the official repo locally with the gradient UI which has also implemented video input and the results are a lot better. I know they still rely on insight face but I would love to hear your opinion about the differences
the jitter got way worse.
Where's the download for the workflow with the changes you made??
I tired following your guide and watched all the beginner videos you linked (since I started from zero) but how do I get my webcam to work with the live webcam example? There is no node to choose my webcam source or something.
there is no manager button in the latest download from github. how i do this?
Can we do something like this fast enough to be interactive?
hi, basically, i had used ur workflow of liveportrait for webcam, but there is an error , Cannot execute because a node is missing the class_type property.: Node ID '#205', so, may u know what custom nodes were missing?
The problem with all these local versions, is their installers and interfaces are usually... abysmal.
Like the programs just got it running, and then released it all unpolished and with a long drawn out nasty install process full of dependencies and other library cruftware you have to install along the way just to get it running.
Agreed.
At least we're not stuck in a command window
The pros of the local versions is they are more often than not, not stupldly censored, and what they can do today for free for youthey should do for a long time to come. A lot of the online AI tools are free to use not in a limited capacity while they are rightly so in beta (because the results you get with AI is more often miss than hit)... but when they do go pay, you realize, they want yearly subscriptions, and it's in no way worth it... to use the same beta crppy tool you were using before, that did not improve, they just finally finished the subscription billing code.
So I'm all for pushing for more open source AI stuff, but the developers really need to focus on polishing up their installers.
@@choppergirl I agree with all of that. Also the 'credits' model I don't like. As you said, most results are miss not hit. So you burn through credits trying to get something useful.
I wrote an installer and then they wrote their own choosing not to commit mine.
Hey Nerdy, on video2video the results are better if you use the same fps number in each video
Yup, you can set that in the “fps” node!
@@NerdyRodent kudos for create amazing insteuctional videos Nerdy
Nerdy-sensei, a good post-processing tool would be facefusion since it can lip sync better, use GPEN for face improvement and also has frame enhancer. Love your work
Yup, if you’re ok with the non-commercial nature of insight face then it’s great!
@@NerdyRodent I forgot they use insightface, that thing is like a virus everywhere. Thanks for the explanation
Where's the Face-Alignment workflow? The examples folder only has workflows with MediaPipe and InsightFace, and there's no Face-Alignment node in the manager.
Nice toy, but not yet ready
2 papers down the line… 😉
@@NerdyRodent papers are worthless, hands on software is all that matters.
@@FusionDeveloper Ok edgelord
@@AB-wf8ek Anyone could claim "papers" without actually having anything real.
I don't know if that actually happens, but I do know that I've seen soo many "vapor-ware" announcements on so-called amazing "papers" that I unsubscribed and blocked "2 minute papers" because it was a complete waste of my time. "hey look at this cool thing...... ... that nobody will ever get to use" over and over and over... It's getting excited for empty promises.
I could CGI render something and make outrageous claims about instant/realtime stuff and claim "2 more papers down the line, you will be really shocked" and just leave it at that.
Fool me once...
Is there anyway to prevent the head pulsing like that?
+1
Haha i like the outro music, how do you make it?
can you share your workflow modify? im newby
Wow that is amazing news!
Finally --- a replacement for InsightFace
I've found facefusion is better but might need to try this what is the insight face model res is it also 128x128?
"Moo moo moo" indeed!
Where is workflow?
The best PC spec for this?
you can do it on a 1080? probably with a ok cpu
It's all about the future. ;-)
this work in real-time or only record?
Only the webcam option works in real time
@@NerdyRodent how its work? webcam enabled but only rec button i see
@@MitrichDX You can auto queue, like in the video
You don't need comfy UI to run LivePortrait.. please correct your initial comment...
The one I highlight is the _only_ open source option and you _do_ need ComfyUI to run it 👍🏼 You can correct your comment now 😉
yay colors!
it still cant do the pumpkin test and fails on many anime faces like the joker
Kolors vs Midjourney?
Great tutorial, Nerdy! Thanks for sharing! 🌟
thanks for all your vids and updates !
Are you single? I love your accent.
Man, I love AI lmfao
video to video, wao
honestly still not good enough to use
Man, the head bubbling is terrible….not ready for prime time……
Is it just me or with every update this tool is becoming worse and worst?
How so?
But live portrait can do 600 frame only 😂
comfy? nope.
I thought I was the only one. lol
Comfy: tool for people with brains wired like spaghetti. As if those 'nodes' were anything else than simple function calls with parameters that would be way more readable - and much better to debug - in code form instead of this mess of a GUI.
And in fact 'install missing custom nodes' too often fails to work.
What non-programmers don't realize is, that moving nodes around like in a kids games has no advantages compared to having code with definite positions from top to bottom. The later being way more transparent and more easy to follow. And easily searchable and replaceable at that.
@@joechip4822 Sadly Comfy and Stability AI are worried that the regular normal person in the street will be able to make content that rivals bug studios, they needed to put a stop to this. That's why thy have been pushing Comfy UI which is only aimed at a minority of people with understanding of nodes. This keeps takes it away from the regular people who helped build and test the ai and back into the hands of the big studios.
@joechip4822 I'm not a programmer, but I am a digital artist.
Node based interfaces have been around for decades. Modular rack synth systems are the original manifestations. You can also find node based editing in all the most advanced software for 3D animation, video compositing, video effects, and music.
Being able to visually patch input/output connections is profoundly more efficient for editing non-linear workflows and complex custom processes.
It's perfectly valid to say you don't like it, but to claim it's not a good system altogether simply disregards everyone who uses it.
It's not like there aren't other options. Fooocus and Invoke seem like great alternatives.
@@AB-wf8ek okok, but do you know what had been around also for quite some time? Wire routing for electronic chips and PCBs - which by principle avoids overlapping wires/cables. If comfy had such an option it would reduce the messy 'everybody does it their own way', 'everything is over the place' and 'I need the biggest screen ever to grok this' approach of comfy. A GUI like this simply needs much more tool functions.
And by the way... What makes me suspicious is, that newbies and seasoned AI artists alike very often despise such node based GUI - reddit has many debates about this, and you can even find the opinion that these systems are the equivalent of LINUX in their ways of making easy things complicated while mainly allowing the da boys to boast about their own ingenuity.
The n.b.d. may come from a time when things were totally analog and machines had actual plugs. But working with a famous modern software synt named 'Pigments' I know that on a digital screen there are better ways, if you don't want to work with code, and the many tools that exist to make this effective.
But granted... It may be mainly a matter of taste and experience in a certain field. Especially artists are very often extremely original in their art products, but cling tightly to a certain limited tool set, even if it hadn't been optimized for quite a long time (Photoshop, Ableton Live etc.)