Control AI Actors with this Awesome Workflow (LivePortrait Tutorial)

Поділитися
Вставка
  • Опубліковано 23 лис 2024

КОМЕНТАРІ • 150

  • @fraterseamus
    @fraterseamus 2 місяці тому +11

    Best tutorial on Live Portrait I've seen yet, thanks for posting.

    • @nikgrid
      @nikgrid 2 місяці тому

      @@DrZaious Agreed. You can do this for free using comfyUI and Live portrait workflow.

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      Wow, thanks!

  • @animateclay
    @animateclay 2 місяці тому +3

    Thanks for the tutorial. it's the most simple solution out there for animating faces that I've seen!

  • @AEFox
    @AEFox 2 місяці тому +5

    For better results, I think it's better that you don't put in your prompt any "talking" "Saying" (mouth/lips movements) anything for the base video, so the mouth and face do not be doing any expression as base face video, so then will be modified from a "blank default/idle expression" of the face by the face controlled video, so the modification will happen to the same face and not a face that it's already doing/changing over time.

  • @IAmVo
    @IAmVo 2 місяці тому +1

    THANK YOU!!! When you opened Comfy I almost clicked away. Its so so complex and I just wanna hire people to collab with at that point. But you broke it down SO well, I understood and it made so much more accessible thanks

  • @heartshinemusic
    @heartshinemusic Місяць тому

    Thanks for the tutorial. These lip-sync features are so important to advance the control over consistent characters. I hope more A.I. video platforms will implement this under the hood, without all the incredible complex Comfy parameters.

  • @DigitalAI_Francky
    @DigitalAI_Francky 2 місяці тому +1

    With the local version and adding an upscale node, you should be able to have it in a bigger resolution 😊

  • @Hadar360
    @Hadar360 2 місяці тому

    Glad that I found your channel and tutorials, great stuff

  • @WattleburyWoods
    @WattleburyWoods 2 місяці тому +2

    That sweater on that background is ❤‍🔥

  • @mitchdavis5159
    @mitchdavis5159 Місяць тому

    Thank you, Curious Refuge!

  • @malusjoon
    @malusjoon 2 місяці тому +1

    i would love to see your studio...how you setup and how you get the ring animation in the background! 🙂

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      Maybe we'll do a desk breakdown someday :)

  • @justinlloyd3
    @justinlloyd3 2 місяці тому +1

    Great tutorial! Thanks you!

  • @Art-ifishl_Intelligence
    @Art-ifishl_Intelligence 2 місяці тому +2

    You were the best sweaters lol

  •  2 місяці тому +1

    Thanks for the info, some useful tips. Wondering how do you get around videos that need lip syncing?
    Also, where can I buy a sweater like yours? :)

  • @ZedMagnet
    @ZedMagnet Місяць тому

    I respect the flexibility of ComfyUI, but only super geeks are going to even attempt to grok it. Higher level tools (even if Comfy is under the hood) will be needed for wide spread usage, IMO.
    Awesome presentation. You are top notch, sir.

  • @gumvue.studio
    @gumvue.studio 2 місяці тому

    thank you, this is good to try . I upload some AI films, 8 min, but will try this way also

  • @FilmSpook
    @FilmSpook 2 місяці тому

    Many Thanks for all your help, Bro.

  • @joakimmogren1727
    @joakimmogren1727 2 місяці тому +4

    I'm a horrible actor and I don't want to record myself. I know this is going to sound incredibly lazy but I hope the option to prompt different emotions becomes available. I don't need Oscar worthy performances for the projects that I'm working on. It's great that's there's a lot option for controlling faces. Now only if we could have 3D poseable consistent characters we'll finally be able to have AI actors?

    • @SykoActiveStudios
      @SykoActiveStudios 2 місяці тому +2

      Metahuman with unreal.

    • @JulioMacarena
      @JulioMacarena 2 місяці тому +1

      Hedra AI - check it out.

    • @JulioMacarena
      @JulioMacarena 2 місяці тому

      ​@SykoActiveStudios thanks. Tell me more. :)

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      You can certainly try other things like Hedra!

  • @LaurenceBricker
    @LaurenceBricker 2 місяці тому

    Thank you! Can you add how to deal with the actual audio in the performance?

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      You'll have to bring it into your editor :)

  • @Sean43322
    @Sean43322 2 місяці тому

    Thanks for tutorial. You used close-up image and the result was almost good and acceptable. My question is, can the same result be obtained with medium or wide shot images? Have you ever tried it?

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      We will have to do some testing! But close up is always best results

  • @abbasmedina786
    @abbasmedina786 2 місяці тому +3

    Awesome

  • @IlluminousThoughts
    @IlluminousThoughts 2 місяці тому +1

    this is really cool

  • @chariots8x230
    @chariots8x230 2 місяці тому

    Can you also do a video about ‘Crew AI’? I think ‘Crew AI’ seems like a useful AI tool because it gives users a lot of control over their AI character’s facial expressions, just by moving some sliders around. It doesn’t really create animations, but it can change the facial expressions of characters in still images. I need to work with both still images and animations, so I like that there’s a tool that can modify characters in still images like this. Also, in terms of animation, I can see a tool like this being useful for creating a ‘start frame’ and ‘end frame’ for an AI animation.

  • @Realidyne
    @Realidyne 2 місяці тому +4

    So if you wanted to sync up with audio, I'd imagine you'd have to slow the final output by 25%? If that's the case you'd have to increase the fps by 25% (around 30fps) if your final final output is 24fps with synced video?

    • @phu320
      @phu320 2 місяці тому

      I just went through a nightmare like this yesterday.

    • @EmilyNilsen
      @EmilyNilsen 2 місяці тому +1

      I found that as long as I had the same frame rate on each video, I didn't need to increase the speed of the driving video by 125% initially. Also, I got better results, prompting for subtle head movements only and without prompting the character to speak into the camera.

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      Would def take some finesse in the editor!

    • @logan_smith
      @logan_smith 2 місяці тому +1

      @@EmilyNilsen Yup. Grab the fps from the driving video via the Video Info VHS node and plug that into the Video Combine VHS node. You shouldnt have to do any speed changing.

  • @okgo5152
    @okgo5152 2 місяці тому

    Thanks, great video!

  • @TheCrackingSpark
    @TheCrackingSpark Місяць тому

    Hey! Great video. I have an unrelated question, kind of an odd one. What's your background? Are you using a greenscreen? If so why can't I see any green in your glasses? Is it not a green screen and just a straight up tv screen? Your shot is lit very nicely, and it matches the background very well! I just had that question lol

    • @curiousrefuge
      @curiousrefuge  Місяць тому

      It's a projector :) Glad you enjoy the look!

  • @julbombning4204
    @julbombning4204 2 місяці тому +2

    Hey I want to buy your sweater asap, where can I buy it?

  • @TheJereld
    @TheJereld 2 місяці тому +3

    I like the new background!!!

  • @High-Tech-Geek
    @High-Tech-Geek 2 місяці тому +1

    Just to clarify if I've got this correct:
    1. If you're uploading a control video to animate an image, you can move your head all over the place and the resulting video will mimic those movements.
    2. But if you're uploading a control video to animate a video, you need to keep your head still as it will only apply your face to the video character that is already moving their head.
    Is it correct to say the control video head movements will not affect the existing video character's head movements?

  • @festivitycat
    @festivitycat 2 місяці тому

    Love that sweater!

  • @Escelce
    @Escelce 2 місяці тому

    Great video thank you

  • @jzwadlo
    @jzwadlo 2 місяці тому +1

    would it not make more sense to also record the reference footage in front of a white wall or plain background vs clutter in the background? Surely that makes a difference?

    • @curiousrefuge
      @curiousrefuge  2 місяці тому +1

      It would likely help! But not totally necessary

    • @jzwadlo
      @jzwadlo 2 місяці тому

      @@curiousrefuge fair enough! 🤝

  • @ai.ai.captain
    @ai.ai.captain 2 місяці тому +1

    ❤❤❤

  • @0A01amir
    @0A01amir 2 місяці тому

    Any site with face portrait videos that we can download ?

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      You can use Midjourney to make thme!

  • @JoJoAIventures
    @JoJoAIventures 2 місяці тому

    Thank you, ComfyUI is pretty much a game changer these days. Every serious AI video creator should consider learning it.

  • @StéphaneLévy-i7g
    @StéphaneLévy-i7g Місяць тому

    Can't get this to work. I've followed the tutorial step by step, and used the specified assets but the video combine node always appears EMPTY and I cannot render. How to fix this? Thanks in advance.

    • @curiousrefuge
      @curiousrefuge  Місяць тому +1

      Feel free to jump in our discord and we can try and help you troubleshoot!

    • @StéphaneLévy-i7g
      @StéphaneLévy-i7g Місяць тому

      @@curiousrefuge Thanks.

  • @stevey19861
    @stevey19861 2 місяці тому

    What background are you using?

  • @photogeneze
    @photogeneze 14 днів тому +1

    why there is no workflow downloadable?

    • @curiousrefuge
      @curiousrefuge  11 днів тому

      It's not something to download!

    • @photogeneze
      @photogeneze 9 днів тому

      ​@@curiousrefuge i mean the comfyui's workflow, simple json file, of course the models and everything else is separate download, i understand that. But why not workflow file?

  • @NoFaithNoPain
    @NoFaithNoPain 26 днів тому

    What we need is faces at an angle and people who are moving

  • @mitchdavis5159
    @mitchdavis5159 Місяць тому

    How do I get the output video to be longer than 8s?

    • @curiousrefuge
      @curiousrefuge  Місяць тому

      Editing software and splice clips together :)

  • @SmartVideoBE
    @SmartVideoBE 2 місяці тому

    can we run it locally on mac or windows without paying? how is this done?

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      Yes but we prefer running it online as it takes quite a bit of cpu power

  • @elgodric
    @elgodric Місяць тому

    What's the GPU requirement to run Live-portrait locally?

    • @curiousrefuge
      @curiousrefuge  Місяць тому +1

      Good question...we'll have to check!

  • @DaveandhisDeathbeanie
    @DaveandhisDeathbeanie 2 місяці тому +1

    Why are people so into the sweater?!

  • @shireensingh2834
    @shireensingh2834 2 місяці тому +1

    Wowwwwwwwwwwwww

  • @khalil2099
    @khalil2099 2 місяці тому

    can we only do lipsync ? meaning, the app will ignore the eye.

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      It's not as refined for other facial features

  • @DanielSierra-l1u
    @DanielSierra-l1u 18 днів тому

    Why does my output come out so damn blurry? Its so blurry i can barely see lip movement

    • @curiousrefuge
      @curiousrefuge  16 днів тому

      Let us know in our discord and we'll try to help!

  • @JulioMacarena
    @JulioMacarena 2 місяці тому

    Tell the actor that you're going to use their single performance many times in different projects before you record them. :)

  • @faisaledits1
    @faisaledits1 2 місяці тому

    Gen 3 Alpha is not free 😔

  • @julx97
    @julx97 2 місяці тому +6

    i dont like ComfyUI, that's why I like to use LivePortrait in "LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Control".

    • @adarwinterdror7245
      @adarwinterdror7245 2 місяці тому +3

      What? What are you talking about?
      Is there an alternative to do what was shown in this video without comfy UI and doesn't involve installing Live portrait on the PC? (Which I had issues with)

    • @julx97
      @julx97 2 місяці тому

      @@adarwinterdror7245 aitrepreneur made a tutorial about this topic. It’s his second most recent video

    • @julx97
      @julx97 2 місяці тому

      @@adarwinterdror7245 a it re pr en eu r made a tutorial about this topic. It’s his second most recent video.
      I had to write spaces in between. As my comment disappears when writing his channel name

    • @findthatvid-pro
      @findthatvid-pro 2 місяці тому

      @@adarwinterdror7245 Find the link in my post (as UA-cam comments get deleted all the time)
      The post will expire in 24 hours

    • @findthatvid-pro
      @findthatvid-pro 2 місяці тому

      @@adarwinterdror7245find the name in my post

  • @GaryMillyz
    @GaryMillyz 2 місяці тому

    AI gfs are gonna be f'in wild

  • @jonathankr
    @jonathankr 2 місяці тому

    Im not sure if my scrept needs that. Lol

  • @Steger13
    @Steger13 2 місяці тому +1

    Still not perfect though need another 6 months to a year i think

  • @deepdiver849
    @deepdiver849 2 місяці тому

    Its not working for me when i upload an image of an ai created cat. It doesn’t recognize it.... so ai is still behind

    • @curiousrefuge
      @curiousrefuge  2 місяці тому +1

      Hmm, jump in our discord and we can try to help!

  • @woltergeist9175
    @woltergeist9175 2 місяці тому +1

    AI is really good at generating non-ugly women

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      True! It's trained on a bunch of models it seems for all genders!

  • @kamizama9177
    @kamizama9177 2 місяці тому +13

    Not easy at all lol

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      We appreciate you trying!

    • @ryan18462
      @ryan18462 Місяць тому

      Every AI tool is the best!

    • @jeremyleonbarlow
      @jeremyleonbarlow 6 днів тому

      If you want easy, Runway has Act-One for talking heads. This however looks like it might be beneficial if you were doing a walk and talk video from an image to video and you needed a better facial performance on the character. Is it easy? No. Is it less expensive and easier than hiring a location, hiring actors, and doing a full fledged on location shoot to get essentially a similar shot? Yeah, and what's more you can create the perfect version of your character if you iterate to get what you want in the original image generation and take the time to train proper LORAs.

  • @robertdouble559
    @robertdouble559 2 місяці тому

    Very rubbery. No regard for underlying bone or muscle structure. But hey, its early days, its only gonna get better. Good enough for memes and amateur work, I guess.

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      It's certainly still developing!

    • @lefourbe5596
      @lefourbe5596 Місяць тому

      Lora dataset augmentation. Great for that

  • @fabianoperes2155
    @fabianoperes2155 2 місяці тому

    Looks too artificial, maybe in w few next versions they will fix it.

  • @雪鷹魚英語培訓的領航
    @雪鷹魚英語培訓的領航 2 місяці тому

    Still not good enough for my taste, but one or two more papers down the line, or if the EMO paper gets put to use I'm game.

  • @ToneVirtue
    @ToneVirtue 2 місяці тому

    I think it is over. All music and film was about relating to fellow human beings. This AI is making everything oversaturated and quite frankly pointless. Doesn't make me feel like watching anything. It is just a pile of plastic straws at this point.

    • @themightyflog
      @themightyflog 2 місяці тому

      Nah. That was done away with when it costs so much to create a vision. Money made us lose the connection.

    • @curiousrefuge
      @curiousrefuge  2 місяці тому

      Before AI, how did you feel about online content?

    • @ToneVirtue
      @ToneVirtue 2 місяці тому

      @@curiousrefuge Great. It opened a market place with loads of opportunities and jobs for people. If you want people to use AI to replace photographers, filmmakers, writers musicians then It will basically kill the whole collaborative market. AI content was not made to improve anything but rather to destroy content creation economy and alienate people form each other. Soon AI won't even need people to write a prompt. I personally think they should ban AI generated content monetization to save content creator economy.

  • @anthonymark2516
    @anthonymark2516 2 місяці тому

    Not bad!
    But my technique is better. As I don't need any input videos, just my imagination.
    And I can get multiple performances, so I can actually direct the actors by choosing the performance that works best.
    /watch?v=iMXbSXSJqqs
    We're at the point now where the tech does not matter so much as how we use it.

  • @Pure_Science_and_Technology
    @Pure_Science_and_Technology 2 місяці тому

    This is bad.

  • @seencapone
    @seencapone Місяць тому

    It might have been better if you would've laid your original vocals over the final clip so we could see how well the lip sync holds up through the processing. No one wants to go through this whole workflow just to make a silent film of people flappin' their lips.
    This reminds me of all the face capture demos -- and I've watched 'em all -- no one demonstrating the tech ever records any dialogue or normal scenes of simple talking, they just pull crazy funny faces like that's supposed to be useful.

    • @curiousrefuge
      @curiousrefuge  Місяць тому

      You've watched them all? We've seen quite a few of people using their real voice (and typically adding some flavor with elevenalbs with v2v)