How to use Temporal-Kit and Controlnet in Automatic 1111 - Incredible results - Tutorial
ฝัง
- เผยแพร่เมื่อ 29 เม.ย. 2023
- In this video I will show you how to use #controlnet with #AUTOMATIC1111 and #temporalkit.
ControlNet : neon.ly/NL8g0
Controlnet Tutorial: neon.ly/vEgBO
Ebsynth: ebsynth.com/
Ciara Rowles: neon.ly/v7JnA
Original Video 📷 neon.ly/voqda
Enjoy folks ;)
#stablediffusion #tutorial - ภาพยนตร์และแอนิเมชัน
Just passing by, but as someone who is currently playing with Temporal Kit on A1111, wanna share with you a few things :
1) Limit your prompt to be as short as possible, because if you put too many "demands" in there, the images/videos will try to change accordingly but not consistent
2) You need a higher CFG (10 or 15) than usual (7) for more accurate results for short prompt
3) Don't use "Guess mode", it will mess with images' consistency.
4) Canny is not accurate try to use Lineart (Realistic).
4) At 4:39 put the main folders path, not subfolders ( 0, 1...), and do all the Ebsyth process first, then it will help you Combine all subfolders at once and merge all crossfade video files into one full video ! (no need to combine one after another)
Thank you for this!
@@karim_yourself ur welcome , bro!
4) Which model should we use with Lineart(realistic)? Also would share a link of your checkpoint
@@mertoncu740 Lineart should work with most of the models. Im using various models downloaded from Civitai.
@@The_Art_of_AI_888 thanks a lot
And we have the honour to have Trump explain us temporalkit, amazing
crisp and short. for people finding the video fast, they can slowdown or pause. dont want lengthy fills with craps and getting lost. thanks man!
Glad you found it helpful :)
I was initially confused but also encouraged to immediately try this just because the video is so short. It gives the impression of VERY EASY TO DO. I really love how compact and concise this video is. It's like a tl;dr version of a long ass tutorial of how to do stuff. I didn't know about how to install tempkit on Auto1111 but ofc I could look for by myself, ebsynth too. Then when everything was done, I just came back here and watch magic happens. Ty so much for this super fast tutorial. One thing I don't understand still, is that my output always save the two preprocessed images from the controlnet (from the open pose and canny), I need to delete them by myself from the folder. Where is the setting to make them not save those images too?
you're welcome :)
a PERFECT way to learn something, thank you!
you're welcome
This is amazing and hilarious
thanks :D
Man, that was a fantastic run down on a really powerful process. Thanks. Just a side note, has anyone ever told you that the cadence with which you speak is remarkably like Donald Trump. Are you both from the same part of the States?
Thanks mate!
Well that's because this is Trump's voice generated by AI 😂 If he's not offensive people don't recognize him, good ear my friend
@@karim_yourself what did you use to generate the voice?
Thanks Donald, the greatest tutorial on TH-cam.
confused, surely changing sides to 1 means keyframes are different and consistency is sacrificed?
thx donald
hi tyvm for the nice tutorial! I get this error, what does it mean?
, line 86, in __init__
'-r', '%.02f' % fps,
TypeError: must be real number, not NoneType
This seems like something i want to do, but after watching your video, i have no idea whats happening. You should start the video with a before and after, so viewer know what the general idea is before diving into the step by step. Subscribed!
püoint taken, thank you
Thanks! Looks like Trump found a new hobby 😂
Super annoying that voice, has to stop watching unfortunately because of it.
Anything for backgrounds? This Process dont do well with backgrounds...
Hi, I've tried this. It does looks very smooth. But i have one problem. When I put all the crossfade videos together, i noticed there's huge frames skip in between those cuts. It should be a continuous smooth one shot. Any idea what causing this?
This could be due to too little keys for ebsy th to create a smooth video. Try to do more keys per keyframe, let me know if it worked :)
@@karim_yourself Thank you for the tips!! I did remember I changed the fps to 24, rather than keeping the default 30fps
找到每4秒卡一下的原因了，我做第一步分割帧数的时候发现，它第0and19和1and1会重复图片，1and19和2and1会重复图片，不知道怎么解决
since when is ebsynth temporalkit?
Each time I open automatic 1111 it reinstall everithing again. Is there any option to avoid this? I use SD on October and I didn't have this easiues.
Same here. If you look at terminal window while A1111 is launching, you'll see that it is set to always reinstall components that have newer versions, or their dependencies have newer versions too. Which often results in installing a lot of components, everytime it launches. Not sure whether it's something I'd like to SOLVE, or just leave it as it is so I don't break something. :)
0.23 - 0.28 is best
thanks, this not works with google drive? so I need to download all and run on my pc after I got the image on A1111?
I have everything local, are you using google collab?
@@karim_yourself yes, 100GB space
video_stream = next((stream for stream in video_info['streams'] if stream['codec_type'] == 'video'), None)
TypeError: 'NoneType' object is not subscriptable
What does that mean?
I can't get past the first 6 seconds, I assume at some point you try to sell us something that comes with free steak knives if we act now, but wait, there's more?
I have a newbie question, is there a tutorial for installing temporalkit on a Mac? Not sure if it's possible. thanks!
Hey Eric, I don't know the answer to this, sorry 😞
ty donald
you're welcome ;)
video_stream = next((stream for stream in video_info['streams'] if stream['codec_type'] == 'video'), None)
TypeError: 'NoneType' object is not subscriptable
I can't understand what the problem is
It means you don't have ffmpeg installed.
Google colab link please 😍😍
can you change clothes?
bro you are talented keep continue
thanks :D
Google colab link please..🙂🙂
You have a problem that nobody is talking about in TH-cam. Controlnet can’t keep up the position of the gaze or eyes……no matter what you do, there is no way to fix this incredible substantial issue……
While the result is good but the process is extremely long. It took more than three hours (with restarts) to convert a 20 second video. I'm sure there are quicker way around
Hey thanks for the tutorial. But honestly, I don't understand the point of temporalKit and EBsynth. from what I understand (and I might not understand) the whole point of the first process in EBsynth and temporalKit is just to break the video into frames. assuming I have all the frames already (let's I rendered a PNG sequence in blender), what is the use of that first process?
Also, i don't get what is the point of doing all that, if you do the same video, a Japanese woman opening a katana on red background, I mean sure, she looks a bit different, but generally it's almost the same video?!
What am I missing?
These are excellent points, I also didn't understand why using Temporal Kit except for breaking video into images. Like yourself, also render image sequences from Blender, and then -depending on how fast your movements are- you can only img2img+ControlNet process every nth rendered image to make keyframes (usually from every 6th to every 12th), and EBSynth using the entire render seuqnce and only those several AI-processed keyframes. Not perfect though as you can often notice the rhythm of keyframes in the final video processed by EBSynth.
@Dr Formalyst yes exactly, I was looking for a temporal kit tutorial because I thought the whole point of it would be to blend and smooth between the img2img frames rendered in batch mode. But I guess not 🤷
@@drformalyst5222 funny enough inside control Net there is a checkbox (forgot its name some sort of frame blending) at the bottom, but experimenting with it didn't produce any better results
@@-Belshazzar- Actually this video explains really well what Temporal Kit does and why: th-cam.com/video/rlfhv0gRAF4/w-d-xo.html
@@drformalyst5222 thanks man, it is indeed a great tutorial
guessmode is now controlnet is more important. I had better results in a dancing video with lineart.
Can you give the github link
For Temporal Kit you mean?
github.com/CiaraStrawberry/TemporalKit
Are these tools free?
yes they are :)
Sounds like Donald Trump teaching AI tutorial here... lolz
Are all of this free? can you guys atleast tell the most important thing.
yes they are free
Wait is this an elevenlabs voice trained on Trump? :)
Exactly 😊
@@karim_yourself 😂 10/10
Gotta say, that's extraordinarily tedious and lengthy. Surely there must be a way to circumvent having to do that over again 4 times? As in doing it via one run-through IF one's got a powerful enough machine?
BUT, I do love your tute approach . You just need to slow it down just a tad and let us know when you're actually pressing one button or the other...EVEN if you say it really quick. It'd still help.
thanks, Trump.
you're welcome ;)
Found it eerily ironic that Donald Trump taught me how to use AI tools
Tremendous if you ask me 😂
Or I simply ditch this bs extension and use an RIFE AI-Model and get the same output in merely a few seconds.
Yeah, I really got frustrated trying to use temporal kit and your video didn't really help me as I already had all key frames prepared.
Is this your voice or is it this Trump text to speech ai generated voice?
The latter 😁
@@karim_yourself The quality if amazing! did you use ElevenLabs to generate this?
thank you donald trump
you're welcome ;)
ten twenty four times ten twenty four on my three thousand and seventy tee