ive been just getting into this and I've been thinking ideas like this and its awesome to see that these techniques do work. cant wait to try this out on my own.
Great stuff as usual bro! I would love to see some tutorials on how to use Stable Diffusion on a Mac, or if you already have one out could you link me to it? The only solution I've seen so far is DiffussionBee which isn't all that great.
Very clever technique tracking an already stylized face over the footage, does it work with faces that show bigger expressions (lauging, crying, going from neutral to smiling etc)?
not a perfect system, but great idea :D excited to see this evolve more. What the corridor peeps did was include deepfake, i wonder if they created a stable diffused face with enough iterations to train deepfake mask and then deepfaked original or deepfaked the actor as Tom Holland and then diffused a stabilized video. thoughts?
They just face tracked and placed it on the original footage, essentially rendering the face differently and the rest of the scene differently. Preety much what enigmatice did here
just a suggestion, you should check out nerdy rodent face thing that he did and combine that with your method maybe that will make the face look more alive and natural. thanks for the lesson
I‘ve seen a few of his videos, they have taught me a lot. I think its possible to make it look alive with this method as well, I just didn‘t go that deep with this yet. I‘ve seen what AE face tool can do, and it definitely allows for facial expressions. I‘ll make another video showing that someday!
Guys! Please help me, the photo that I add to the "Replacement Comp - Holder" flies to the upper left corner, and is not put on my head in the video. + + + + + + For some reason, the "Replacement comp 1" folder has a black screen instead of a video.
Yes it could work. There is a way to do it with the AE face tool i mentioned here. But you need to make some adjustments. May make video about in future.
a better way to get prportion was to do an img to img with spiderverse SD model on your subject and then use that image as a comp instead of gwen stacy image.'
True, however to get a very consistent looking cartoon style that retains same facial features through out is almost impossible. I tried it myself and I was able to get like 1-2 seconds of consistency, then suddenly the style started to shift as her face started to move direction.
@@enigmatic_e I was talking about following the exact same procedure as you did in this tutorial but i stead of using gwen stacy face from the spiderverse movie, use A spiderverse SD model img2img on your subject. Than use that generated image instead of gwem stacy one and follow the same process. This will reduce all the headache of "getting the same proportions"
Hello bro, Your videos are informative and amazing I'm using RTX 3060 and still having trouble calling ( CUDA out of memory. Tried to allocate 678.00 MiB (GPU 0; 6.00 GiB total capacity; 4.65 GiB already allocated; 0 bytes free; 4.81 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF) Please let me know if you know the solution
Hi, comments from an old audience: maybe I can call you "master", a master in AIGC field. I hope you can continue to lead us to explore more interesting and useful AI usage. I will always subscribe and recommend you to the people around me.
@@enigmatic_e damnn.. actually is so good that I can't believe it was just that hahah. Can you do a tutorial? This all final look would be a good tutorial.
Watching the steps.. opening Snap Chat, turning on the Disney cartoon filter.. Voilaa mission accomplished lol.
Bro, you are in the cutting edge for fine tuning the video output of these AI based platforms. Keep it up!
Slowly but surely, you are creating the ai workflow of the future !
Great tutorial and an awesome technique. Thanks for this.
Many thanks mate ! Subscribed
you da man! thanks a million!
dude this is absolutely amazing! Thx a lot for sharing
Such a cool process! Thanks for taking us through your creative process, its really amazing!
ive been just getting into this and I've been thinking ideas like this and its awesome to see that these techniques do work. cant wait to try this out on my own.
dude your videos are wesome you dserve more followers!
Thank you!
Consistent!
great work !! congrats !!
Bro, thank you! Great work!
Always great content my friend ;)
Thank you!
I appreciate your Amazing work, I'm learning alot from your channel
Happy to hear that! 😊
i would look into ai background removal!! it’s worked really well for me and might save you some hassle with masking
It's free or paid? Can you send me the website?
@@ianhmoll look up robust video matting
Rotobrush 2 is ai based tho
@@jordyandrews thank you. Seems pretty good
Excelente men.
This is nice!
thank you bro!
Great tips !! Thx a lot !
thanks a lot dude ^^
Great stuff as usual bro! I would love to see some tutorials on how to use Stable Diffusion on a Mac, or if you already have one out could you link me to it? The only solution I've seen so far is DiffussionBee which isn't all that great.
😃Big job, Stable Diffusion top. 👍
Too difficult for me but the result like the one you've got motivates to develop skills
corridor crew did the same thing but you made a tutorial about it thanks!
LIKED, SUBSCRIBED.
Very clever technique tracking an already stylized face over the footage, does it work with faces that show bigger expressions (lauging, crying, going from neutral to smiling etc)?
Technically you could with Face Tool. I‘ve seen footage of it being used while people show expressions. I may do a video showing that.
@@enigmatic_e yes please!
not a perfect system, but great idea :D excited to see this evolve more. What the corridor peeps did was include deepfake, i wonder if they created a stable diffused face with enough iterations to train deepfake mask and then deepfaked original or deepfaked the actor as Tom Holland and then diffused a stabilized video. thoughts?
I think that is very possible.
They just face tracked and placed it on the original footage, essentially rendering the face differently and the rest of the scene differently. Preety much what enigmatice did here
i dont think they used deepfake for their spiderverse video
@@kylokat they didn't you are right. they used face track
@@kylokat They termed it a 'stable fake' where the actor playing tom holland had his face replaced by a Spiderverse Tom Holland.
nice idea
Thought: This might actually be easier to do in blender
As a blender user, I was just thinking the same.
Yeah, that might be true.
Nice. Sometimes I use blue/green background then you can key it out after you run it through Stable Diffusion instead of needing a mask
Yea i think thats a great work flow.
Nice one, will definitelly try that aswell. I just wonder if it would be better if you stabilize it to the face first and then apply the cartoon face.
Im sure they both work. You can try and let me know if how you like it.
just a suggestion, you should check out nerdy rodent face thing that he did and combine that with your method maybe that will make the face look more alive and natural. thanks for the lesson
I‘ve seen a few of his videos, they have taught me a lot. I think its possible to make it look alive with this method as well, I just didn‘t go that deep with this yet. I‘ve seen what AE face tool can do, and it definitely allows for facial expressions. I‘ll make another video showing that someday!
It’s not free, but Lockdown on aescripts would save a ton of work in this workflow.
Guys! Please help me, the photo that I add to the "Replacement Comp - Holder" flies to the upper left corner, and is not put on my head in the video. + + + + + + For some reason, the "Replacement comp 1" folder has a black screen instead of a video.
Nice thanks for sharing! So this would work with a talking character just as well?
Yes it could work. There is a way to do it with the AE face tool i mentioned here. But you need to make some adjustments. May make video about in future.
still hard work
Yep. I mean, at least people can‘t say AI videos takes no skill.
I saw that you use inpainting as a template. What other stable diffusion models are good for maintaining consistency?
Now you can use almost any model with ControlNet and get some pretty consistent results.
sir i have a question can we create a vlogging video like this metaverse type like tthis can we create please tell me please
hello, I would like to know how you use this transition in the multi-colored wave video, could you tell me how you do it?
a better way to get prportion was to do an img to img with spiderverse SD model on your subject and then use that image as a comp instead of gwen stacy image.'
True, however to get a very consistent looking cartoon style that retains same facial features through out is almost impossible. I tried it myself and I was able to get like 1-2 seconds of consistency, then suddenly the style started to shift as her face started to move direction.
@@enigmatic_e I was talking about following the exact same procedure as you did in this tutorial but i stead of using gwen stacy face from the spiderverse movie, use A spiderverse SD model img2img on your subject. Than use that generated image instead of gwem stacy one and follow the same process. This will reduce all the headache of "getting the same proportions"
Hello bro,
Your videos are informative and amazing
I'm using RTX 3060 and still having trouble calling ( CUDA out of memory. Tried to allocate 678.00 MiB (GPU 0; 6.00 GiB total capacity; 4.65 GiB already allocated; 0 bytes free; 4.81 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF)
Please let me know if you know the solution
Its too bad they haven't created a quicker way to do this. Hopefully in the near future. :)
Yea i feel you. These are just work arounds for SD limitations. I'm sure very soon, this won't even be an issue.
Hi, comments from an old audience: maybe I can call you "master", a master in AIGC field. I hope you can continue to lead us to explore more interesting and useful AI usage. I will always subscribe and recommend you to the people around me.
Thank you so much. 🙏
Yeah after the video with the stabilized footage i thought the same, now how do i unstabilize and keep the stuff in place, thank you
I know right! No problem.
Cool, ai rotoscoping.
1k like
Niice! How you did this vortex effect on the face transition?
Ah yea, I did that with shape layers and turbulent displacement.
@@enigmatic_e damnn.. actually is so good that I can't believe it was just that hahah. Can you do a tutorial? This all final look would be a good tutorial.
@@ianhmoll yea I will!
@@enigmatic_e waiting anxiously :)
@@enigmatic_e Will you still do the tutorial?
have you tried using dreambooth to improve consistency?
No I haven‘t. Might work to give a specific likeness of a character.
very hard
I know, but sometimes necessary to get what you want.
Can you export this stuff to create a IG filter?
I think theres other ways to accomplish that, probably way easier than this.
@@enigmatic_e auto generated IG filters would be insane 👀
Hmm wonder how this would work with a face that is talking or showing emotion
Use a deepfake tool on the static face with original footage?
The Face Tool allows for emotion and even talking. Will make a video about that some day.
@@enigmatic_e AE plugin?
@@BryanKesler yea, the one i showed here
You could have used "Thin-Plate Spline Motion Model for Image Animation" for better face tracking and moving eyes .
Nerdy Rodent youtuber has very useful tutorial about it
definitely not, thin plate is not that good