Multi-ControlNet - How I got consistent videos!
ฝัง
- เผยแพร่เมื่อ 15 พ.ย. 2024
- Experimenting with Multi-ControlNet settings to get consistent animations.
HOW TO SUPPORT MY CHANNEL
-Support me by joining my Patreon: / enigmatic_e
_________________________________________________________________________
SOCIAL MEDIA
-Join my discord: / discord
-Instagram: / enigmatic_e
-Tik Tok: / enigmatic_e
-Twitter: / 8bit_e
_________________________________________________________________________
Install ControlNet: • Controlnet 1.1 for Sta...
The corridor crew had some pretty good results with keeping that flashing down with a de-flicker plugin. I keep needing to remind myself that I dont have to do everything in SD. I can still use my other tools.
I just saw that. Tested the deflickering and it definitely improved it. Showing results maybe in the next video or in a short! You’re right about using other tools!
Super sick ❤
dude this is amazing, cant wait to try it, thanks for share, thats so solid output
Great vid bro, keep it up!
really like the style of the video. not a typical tutorial. just a dude testing out an idea. subbed
My dude, loved that you see it that way. That’s definitely what my intention is with my videos. I don’t want to come off as I have all the answers, cause I don’t, I’m just messing around trying to figure out what works.
@@enigmatic_e looking forward to more stuff from you. keep it up
See, this is why I love your channel. You're always trying new things and experimenting while staying humble about it. You're the only channel out here working with AI art generators and After Effects. Maybe word still needs to get around that this is possible, but once it does, your channel will blow uuuppppp!!!! Keep up the great work my friend 🤟
Comments like these motivate me to keep pushing. Thank you! 🙏🏽
@@enigmatic_e of course 🙏🏿🤗
Great video, really enjoyed following along with your experiments and seeing what did and didn't work for you. The jks are fun too, keep on keepin it fun, light, and learny.
You are doing a great job experimenting, I am also trying everything I can, I hardly sleep! but it's very exciting!
I feel you on the hardly sleep 😂
Sleep is overrated°°°
Nice job bro, you pretty much nailed the consistency there :)
Awesome! I had ignored the hed model until now. Then I tested it for a comic book purpose and it worked so well. Thanks!
You mean the deflicker? I have a video about it th-cam.com/video/PeXRFH2I9Is/w-d-xo.html
Nice job man! Thank you for this comparison. I would love to see what you could do using EBsynth for some cool animations with realistic output.
Yes me too! I will try it soon!
I can't believe, you only have 23k subs? You deserve a million subs already.
I hope this will be the case someday. Thanks for the support. 🙏🏽🙏🏽🙏🏽
this was really great and helpful;. Thanks Fam!
thats awesome bro thanks for the video ..hugs from Brasil
Great work bro ^^ , i checked the last video of Corridor and the moment i finished it i thought oh yeah , he sure will watch it and maybe ...... idk i just thought maybe you will talk about it or about the techniques used ( the training model technique for example ); keep it up and we are waiting for your next video bro , Much love .
Thanks bro! I saw that vid it was great! I’m sure they wished they had controlnet when they first started. I just tested deflickering and it works very well. I will post results or talk about it in new video.
thank you for this, i'm experiencing the same problem. but this video help me so much!
🚀🚀🚀🎉🎉🎉
Very nice! thanks for the demo!
Thanks for sharing with us
Its definately to see all the tweaks and tuning from others that you would miss in your own sleepless, keyboard bashing, cussing time using SD.
That's cool stuff! Gonna try it myself.
I've been looking into SD and controlnet to see how it could enhance my Stylized 3D animations recently. Your video has definitely helped me get a better understanding of what can be done with these tools. Also just an unrelated sidenote, I would better describe this as Rotoscoping rather than animation.
Super cool 👍
Ok, that is super cool 😎. Thanks for sharing your learned info. ..I wonder if one could do similar with simple drawing flip book images with right seed and settings. Subscribed
👍🏽
workflow is definitely an important part of using stable diffusion
I just ended to follow this video. Bro its great! But I discovered something by accident.
I applied all your settings, except the "normal map", cause I don't have it.
I was using only canny and I got an excelent image since the first try, the funny part is that I tried to apply normal_bae, cause I couldn't find normal_map (Do I have to download it?). And thinking I was appying both tabs, I forgot to "Enable" the "normal" tab and I got a good result XD.
Nice work!
Sorry if anything was confusing. This video is pretty old by AI standards but I'm glad its still helpful.
@@enigmatic_e Not at all! Amazing video. In fact I want to do animation using the corridor method, and I was very happy to get nice results. I'm just confused how to get the "normal_map" or it was changed for another thing? Thanks for answer!
Don't forget to press preview annoator result to see the controlnet preview before hand!
That’s good to know, thank you!
nice work! thanks for sharing
Thanks for watching!
increible!!!! thanks a lot
really good video pal
Thanks!
Awesome . Iv been trying to turn my self into a werewolf while playing guitar but stable diffusion has the video glitching here and there. This just might be what Iv been missing XD. cheers and thanks.
barbaro ta bueno el algoritmo de youtube, 50 minutos y me aparece en recomendaciones!
TH-cam me está echando la mano? Que padre 😂
Stable Diffusion needs to be more stable in other aspect ratios 💯
You can add another controlnet with eg depth and get more consistent results.
I gotta try that!
@@enigmatic_e guess mode can make a big difference as well. It's a lot to get to grips with rn
I actually like the left side more. XD Still freaking mindblowing to be able to do this!
Check out my results here, I was able to remove the jitter th-cam.com/video/PeXRFH2I9Is/w-d-xo.html
Wow… impressive results… I need to get on the Controlnet… I just used my animation degree to make a hat… I heard it gets cold on the bread line… haha keep up the good work…
Try it bro!!
I think I have to after seeing this... I haven't been meddling much with Stable Different recently but I feel controlnet will lure me back in... I try to get out and it just keeps sucking me back in!!! haha
@@iamYork_ 😂
3:03 we've all been there smh. hours of tinkering to fix some basic shit
Different videos have its unique "ideal setting" for consistency, just need to find it out by tring different parameters.
You can click preview instead of Generate for preview cnet 😁
How did u add black background to the original video? Did u mask every frame manually or is there an effect that does that? thanks for the informative video
Hi. I have a question, I have a video in sequences and I took a frame and transformed it with Stable Diffusion and then improved it in Photoshop and got better results. But now I would like this photo created with Stable and Photoshop to follow my video sequence as if it were Ebsynth. And I can't figure out how to do it... Is there a way?
in controlnet tab, choose the option to use your own reference image, and upload your next frame; in the img2img tab, upload the image that you edited. Then, if you tune the denoise well, your generated image will have most of the features that were present in your edited image, but also with the topography and poses of your reference image. Basically, you use the "use your own reference image" functionality to generate the next frame based on the present frame... you start off with the beautifully photoshopped image, and hope that the next images down the process are just as good. If your frames are continuous enough, this should work.
great video, do you know how to use preimere to blend the ebsynth results more smoothly?
Thanks for the video. Are you processing frame by frame? I can't get my head around this new M2M script option
No I just export it through the batch tab
thank you for sharing your workflow, i would really like to know how ebsynth would improve a scene like your in particular
I will try it!
Thank yoy bro!
No problem!
Please feed to ebsynth 😮. And try what the corridor did the deflicker thingy ?
@enigmatic_e did you rotoscope the footage before adding a footage path to SD?
Yes. I then took the generated images into After effects to use the rotoscoped footage as a matte to remove the black background of the Ai generated images
Great stuff "E" !! Thx for the share. Do you know if there's a way to batch a sequence with a precompute Normal or Depth map? From 3D render pass for example.
Mmm I don’t think so, I could be wrong because I’m not too familiar with how precompute works.
@@enigmatic_e Thx for your answer! The passes from 3D software works great with controlnet for image génération, they're similar as the preprocess from controlnet. I guess if it's possible to point a folder with precalculated Normal or Zdepth, and use this sequence in place of the preprocessed one.
Good job 👍
Thanks! 🙏🏽
😂 working so long that there was an update is hilarious
you my good sir are an ai wizard
Please tell me what If I created my own character in photoshop and want my character in different poses , is there is any way to do that. ?
great result for Saitama without CN did you use a Lora or just prompt? he look just like in the anime
I live my life a quarter mile at a time!
😂
Can you post a quick re-do with ControlNet 1.1? A lot of the UI has changed.
Yea I’ll look into it
Is there a way to batch cut the images square? Awesome video thanks!
Not that I know of
Please share a finished video. I believe you will remake some early videos you did with SD now with controlnet.
How many hours did I like to render in CONTROLNET?
about 30 minutes if i remember correctly
@@enigmatic_e Uoolll, very fast, thanks bro
Idk what i'm doing wrong, I cannot get consistent images, ive been going step for step through multiple videos, every frame is a different face and way different articles of clothing. only thing the same is the color of clothes.
10/10
mat1 and mat2 shapes cannot be multiplied (77x768 and 1024x320)
Great video! Spongebob ripped off Ren and Stimpy. Also, Ren and Stimpy is VASTLY superior to Spongebob in every way, shape, and conceivable form. Just want to make that crystal clear🤣👍👍
I totally 💯 percent agree!
could u scroll a little bit slower, i try to find something for video edit but cant stnd what u setup on tabs(((
dont create for youtube algorithm, create to add real value and people will support and u can build a real community. not hating tho much respect
This was very much not a tutorial of any kind. Cool vid, but totally mismatched/clickbaity title. I've been experimenting with animation, controlnet, gif2gif, ebsynth etc. and really trying to nail down removing those inconsistency errors and it's quite a challenge, so i was hoping for some insights here but was left feeling unsatisfied in that regard.
It was interesting watching your workflow, but there was a clear lack of explanation of what you were doing, things were very rushed. Would actually like to see a tutorial at some point though, if you'd consider doing it. I'll tune in.
I didn’t go over the details because I had already made a video previously where I went through what everything does. This was more of a follow up to that. th-cam.com/video/1SIAMGBrtWo/w-d-xo.html
Also thanks for your honesty, I will make note of that for future videos. I’ve had complaints that sometimes I waste time talking about “obvious” things and so I think I may have gone to the other extreme. Again thanks for the feedback.
@@enigmatic_e Ah fair enough. Yeah i still think it's best to be slow and simple, forget the elitist fools who would condemn you for helping others, it's arrogant, not everyone is at their level of understanding, what might be obvious to them is not obvious to others and they should learn some respect and allow you to accommodate for everyone.
Trust me, explain things as if you were talking to a 5 year old, nobody actually wanting to learn will complain about that, they'll appreciate it, it's not patronizing, it's how a good teacher rolls, because if you can't explain something simply enough, then it usually means you probably don't understand well enough yourself.
I'll be checking out some of your other vids and will leave a sub. Best of luck, keep up the good work.
Thanks a lot for your video! I wanted to ask you, how did you manage to get a decent face for your character with a fullbody picture? I struggle a lot without inpating
hi, i am trying to get into the ai animation topic and i have a question, would be cool if you could answer. Why wouldnt you use style transfer on some keyframes of the original footage and then proceed with ebsynth as you did in the older tutorial?
Hi, style transfer is more patterns that follow a style of an image. This is trying to get a person to look like an animated character. Also ebsynth parts best with subtle movement.
Cool! Also, what's the name of the song you're using?
It’s some royalty free music from Envato or Storyblocks can’t remember where it’s from.
@@enigmatic_e Thank you! It was the first song under Chill Hop by Gushito on Storyblocks.
@@JolyLad nice!
plz brother i wasted my 5 hours trying to install control net to my stable diffusion automatic 1111 thingy and i cant please tell or make a video of where i can do that
here is it bro th-cam.com/video/1SIAMGBrtWo/w-d-xo.html
what did u use to get rid of the background on the vid and what vid format is it?
I used some background removing plug in. I go over it in this video th-cam.com/video/Jhuf6M_VrBI/w-d-xo.html I also did some rotoscoping to clean up some areas.
how did you remove the bg?
still trying to get control net to work, non of the files pop up for me when I try and activate them, they were all auto installed into the folder they should be in
This video is outdated. Here’s a newer version Controlnet 1.1 for Stable Diffusion A1111
th-cam.com/video/DIrIL863-LY/w-d-xo.html
@@enigmatic_e I'll check it out! Thanks
I'm not seeing an annotator resolution setting or the canny low and high threshold. Any idea why?
Do you have the latest update? Not sure why you wouldn’t have it
@@enigmatic_e it may be my webui is out of date. Thanks for the response.
incluso el bolso lo mantiene con control net
my stable doesn't have control net options u.u"
Go to the video before this to learn how to install this.
th-cam.com/video/1SIAMGBrtWo/w-d-xo.html
dear enigmatic:
Did you do the photo-by-photo conversion or did you select all the photos and press something in stable diffusion to do it automatically?
I need help getting rid of the flicker! Have you redone this with the lora on?
I use a de flickering plug in by DaVinci studio, there’s a ve on my channel talking about it
@@enigmatic_e i found it and instantly subbed. TY! I'm rendering a new batch now and downloading davinci
@@HogwartsStudy just know it’s not for the free version unfortunately
@@enigmatic_e aww, i thought i saw someone in the comments say it was.... rip :( thanks anyway. I'll stick around, for science.
@@HogwartsStudy yea there should be something coming soon, I’m sure of it. And I’ll make a vid about it
Why not EBsynth this and get more consistency?
I have many videos, some on ebsynth. also this video is outdated already.
I saw the temporal kit one you made but it doesn't stack up to tthis one. I think this one still holds pretty good till controlnets get even better.
I don't have normal map as an option. How do u download it manually if thats what u did?
This is an outdated video I would suggest you check out this, th-cam.com/video/DIrIL863-LY/w-d-xo.html
@@enigmatic_e thank you
With Controlnet installed, there's no need to run the img2img alternative Script anymore?
Not really. Img2img does give a certain style but it typically stay close to the original footage. Controlnet allows you to push the style a bit more.
@@enigmatic_e Thanks for the fast reply! I saw your new workflow video right after I commented 😁
thanks for making this video. im 2:11 minutes in though and i feel like you shouldn’t skim through or go so fast. more detailed and elaborative explanations can be super helpful for beginners or people learning. it feels like ur assuming people know what youre doing which defeats the purpose of a tutorial
the tutorial is very dramatic but many steps are missing to try to do the steps you did to get the result
sorry I like dramatic. 😭
@@enigmatic_e I like drama also. But how can I get from you more information about that tutorial.?
Even if it needs some price. I am ready.
Thanks for your answer
Bruh Lmaoooo
IS THIS AN APP
It’s a browser interface for generating AI art.