Hey guys, welcome to the second episode of this
week in comfyUI - we are February the 18th of 2023 and things have changed again! I'm making this
series because I know it's quite common for people to gauge how far they are with comfy techniques by
watching YouTube videos, but I couldn't find any that summarized what had happened on a week by-
week basis and so much has happened! I'm sorry I missed last week, but in exchange, I've released
four new quick tips and a full-on tutorial, so go check this out on my channel!Thank you! Brought
to you Live from the Alien mothership, here are the latest comfyU Inews. New models, new nodes,
new controlets! Violence, riots, explosions and nukes - the AI apocalypse is upon us, but are you
ready? Psyched! Before we get started, I wanted to point out that you guys have the attention
span of a 5-year-old on ritalin and that none of my videos are monetized, so if you're watching
ads it's not me putting them there okay? So please consider subscribing and leaving it a like if if
you enjoy my content, without it YouTube won't recommend those videos - also I heard it was
seven years bad luck so please. And by the way, uh... those stats are very disturbing - but I
digress. Okay, the first thing we have today is the release of dynamicrafter for comfyUI. So these
are nodes you can install them they run exactly like any other node use comfy manager or install
them manually, it works exactly like animate diff, it has pretty decent results but it also has
limitations, I think the main one is it doesn't have the support and love from the community that
we're getting from Kosikadink and Animatediff, and the second is that it's extraordinarily
slow even on a 4090, so have a look at those, but in my opinion there are better options
for the time being - this may change tomorrow though! Talking about animatediff evolved, it
got updated yet again but then again it gets updated every 12 hours or so, so Kosinkadink is
an absolute machine, and he posted something very interesting on Discord, on the banodoco server,
let me show you - yes it is what you think it is, it's Stable Cascade running in animateDiff! So I
will of course have a full tutorial on that coming up, but until then, uh you can go and have a play
with it! Talking about Discord actually, there is a new Discord - oh this looks very interesting -
what is this - it's all in alien language I wonder who could make something so disturbing, yeah
I created my own Discord because I wanted to touch on new topics. I wanted to specifically talk
about the philosophy of it especially since Sora got released and and it's a server on which I post
things like this, my artwork that's otherwise not directly linked on YouTube. and it has all sorts
of fun stuff including some really cool games by @eerop where you can check if you can guess which
one is the AI generated image or the real photo, so - can you tell the difference between the
two? I couldn't , but maybe you'll do better than me ! Go check it out, I posted the link with
everything else into this description. Going back to comfyUI, specifically models for comfyUI,
it's quite common for people to get stuck with an old version of a model, because they don't
realize how fast they get updated! Look at this, that was two days ago: a new version of
dreamshaper XL, it's absolutely fantastic and it works in turbo mode, it's good for anime, it's
good for photo-realistic, it's good for just about everything! I love dreamshaper, it's brilliant but
there's also another new model helloworld xl 5.0 GPTv4, yes I'll get there eventually... pretty
difficult for me to pronounce all this being French and all - and it's got this really cool
looking photo-realistic feel, it's got specific keywords that you can use that have been listed by
the author, so go check it out! I've created a few images for sourcing to SVD and I had excellent
results with it, I highly recommend it! another excellent model, is copax timeless XL - now that
one is also very recently updated with a V10, and what makes this one really unique in my opinion is
that it's one where you can actually go and create with with DPM ++ 3 SDE, and that's quite rare.
I've created this John Wick frame viewed from the back, I can't show you the ones from the front
or the other shots because my content is PG-13 on YouTube and I would get banned but let's just
say there's some stuff SORA will never be able to! In other news we announced the tragic death
of dragNUWA, oh dragNUWA we barely knew you, it was only last week I remember that I was having
so much fun with that tool, it was implemented in comfy and what you could do is, you could
essentially simulate movement by dragging objects around just like you do in commercial solutions
like Runway. But it looks like something like Microsoft CELA - aka the Microsoft legal hit
team - went and did the laundry, so now you're left with this research paper to remind you of
the good memories you had with dragNUWA. Will it be back? I don't know but. I can tell you
that the fight between free, Libre open source software and Commercial solutions is only going
to accelerate over time. I know which team I'm on and I recommend you do everything in your power
to support free, Libre software through authors Patreons, things of that nature, because we doing
some cool stuff on our local computers aren't we? And we don't want that to go away! Next up we
have another great release by Mat3o, the creator of Ipadapter plus, and that's native support for
instantID within confyUI, and before you get smart with me, yes I understand that doesn't mean it's
written in assembly - of course - but hey don't be a smart-ass with me! Okay, okay, all right ...
so what is instantID? It's not face replacement like some people think, it's ID preservation
across styles, so you can have this picture, for example, photo-realistic - transforming
into anime back and forth, etc. It's brilliant, so go check it out! Absolutely brilliant piece
of work again by Mat3o, he has a YouTube channel, everything is linked in the description, as usual!
Wait... what! oh (dreams do come true) - oh right! I'm really sorry about the comment I just
made about you being a smart-ass, I wanted to say I'm very sorry, I apologize! And that won't
happen again! Next up, we have comfyUI KVM nodes, so these have been recently updated to support
more models, have more features, what do they do? They allow you to use visual models within
comfyUI directly within your workflow - so yes, it means exactly what you think it means! You can
take last week's Moondream, a tool I showed you, which is essentially a model that reads a picture
and spits out some text as to what's in there and you can even oneshot it by asking it who's in the
picture for example. And it's "John Kerry" yeah not the other John, that you may know from this
famous movies, because we don't want any problem with copyrights now, do we? But jokes aside, it's
super useful. It essentially allows you to reverse engineer prompts, and if you're creative with
these tools you can do also do some really cool stuff with regards to in-filling especially with
the recently released foocus inpainting nodes, so go check it out! Another super interesting
release is Motion Director which was previously a standalone tool, allowing you to create your own
motion LORAs for animatediff. They have now been integrated directly into comfyUI by none other
than yes, Kijai, another great developer and this is super exciting, I mean for me as a filmmaker I
love the idea of being able to take my cell phone, uh shoot myself, no, not that way- but you know
shoot myself with a cell phone and essentially capture movement, have my own movement source and
I'm no longer dependent on 3D models, on films, on things of that nature, and being able to take that
movement pass it into comfyUI and not just use a controlnet, but actually capturing that movement
properly and reproducing it like I showed you two weeks ago! So go check it out! Another great set
of nodes. To train it you'll need a 4090 or maybe a 3090, it works about with 16 gig of vRAM but 20+
is better. And now, for those that are still with me, some really exciting new nodes came out last
week: one of them is pro-Post by Digital John, and what that does is allows you to have control
over the image, essentially post-processing within comfyUI itself, without having to open a third
party tool and do the editing there. You can do things like adding film grain, etc. I
think there's like five filters in there, it's quite limited of course, but it's only going
to get better! It's brand new, go check it out, go download it, a really cool set of nodes.
Another great tool for postprocessing that I had no idea was out there are the jovimetrix nodes,
and quite frankly they absolutely blew my mind, because there are so many of them, and you can do
so much stuff with them, it reminds me a lot of confyroll in many respect. I mean if we go look
at a workflow, of course this is a little bit over the top, right, but it reminds me so much of
confyroll, you got so much control over the image, you can do masking, you can do all sorts of
stuff, go check it out, that's another great set of postprocessing tools which should must
have in your arsenal. For something a little bit different, we have face fusion, and I understand
it's not directly related to comfyUI. because it has its own GUI, it's something you install
separately, but I know so many of you depend on Reactor, and I understand that for a lot of you,
it's a big problem, because of the bugs and the crashes - I mean we're all familiar with this -
and as you know, when you're using animatediff to create videos in AI, you need that faceswapper,
otherwise your talent so to speak - your actor - is going to change between scenes. You do
need a face swapper, that's just a fact of life, so the more it gets updated the more face fusion
is becoming the dominant face swapper out there, so go definitely check it out if you haven't
heard of it, it's absolutely excellent! And I kept some of the best for last, that's automatic
CFG, it's a single node it sits between your load checkpoint and your kSampler, and what it will do
is it will automatically adjust your CFG for you, so you can rapidly switch models and it will
automatically find the best CFG every single time. I tried it, it works really great, it's
great for testing stuff and you can chain it with things like freeU_v2 or Deep Shrink, if you're
familiar with nodes, so go check it out. Please, no one ask me for a workflow for this one,
okay? And plus, if you want to get my workflows, you can buy them on Etsy because some weird
guy is selling my stuff over there :( - and I can't do anything about it. Anyways, so never
EVER buy workflows, that's the tip I will leave you on before we see each other next week or
on my Discord. Take care guys, see you later!
