Ай бұрын
Пікірлер
@jonathaningram8157 15 сағат бұрын
why can't sage attention be installed like any other regular app ? it's either WSL or visual studio...
@evolutionsfake Күн бұрын
2:51 really hit me the most in this video... Thank you.
@miguelrodriquez9790 Күн бұрын
great video super helpful
@jcinewilliams8819 Күн бұрын
Trying to follow along not really savvy at all with this kind of stuff but I do have Comfy UI installed I just dont know where to find ComfyUI/custom_nodes?
@cognibuild Күн бұрын
go to the folder that you installed it in.. click on the comfyUI folder.. in that there will be a folder called custom_node :D Feel free to come over to the Discord and hang out with us and learn!
@jcinewilliams8819 Күн бұрын
@@cognibuild Im going to delete it and try it again
@jcinewilliams8819 Күн бұрын
@@cognibuild BTW what is the discord?
@cognibuild Күн бұрын
@@jcinewilliams8819 Discords is a very popular "chat" server where we can communicate and work on projects together. My personal server is here: discord.gg/6r3JnXRD
@cognibuild Күн бұрын
@@jcinewilliams8819 i also have an easy installer for memebers over on my Patreon page if you need further help
@SomeKindOfBears 2 күн бұрын
feels like this guy could do a killer Norm McDonald impersonation.
@אלוןגפן 2 күн бұрын
hey, great video man! Is it possible to train this model for a completely new language? (non Latin alphabet).
@cognibuild 2 күн бұрын
i dont see why not.. you just need to feed it the wav and the text. what an interesting projet. keep me in the loop!
@אלוןגפן Күн бұрын
@thank you for your answer! is there any way you can help me with some information? I have very little knowledge in training models, ive heard that i might have to change something named tokenizer, for it to work
@אלוןגפן 2 күн бұрын
hey, thank you for your amazing work, is it possible to train and finetune F5/ GPT-SOvits V2 for a new language? ( im looking for a good Hebrew tts, for a huge personal project). could really use your help :)
@codyboulware7400 2 күн бұрын
"gag me with a Smurf" lol that's a new one.
@VintageForYou 2 күн бұрын
What about LatentSync In ComfyUI.
@cognibuild 2 күн бұрын
@@VintageForYou working on both that and a stand-alone as we speak. Should have a video in a day
@cognibuild 2 күн бұрын
Its up! Ready for Download.. standalone and ComfyUI ;D
@christophefanjul1224 2 күн бұрын
for mac please
@CozyBrushery 2 күн бұрын
Great
@androjackson480 3 күн бұрын
Yoo the goat is back
@DannySmith-bg1pk 3 күн бұрын
Underrated channel. I wanted your opinion on which among Rope and Face Fusion is better and easier to work with? I saw your tutorials on both and I think you are the best person to ask this. I have the latest version of Rope installed and I just installed FF 3.0.1 but I don't see much of a difference tbh. Would you disagree with me?
@cognibuild 3 күн бұрын
i have opinions on this that are based of both technical and personal reasons (and I've separated them to be fair). Technical: 1) Rope is out of the gate the better software - it can handle multiple faces, which is something FF cannot do 2) Rope has "markers" - which is a POWERFUL way of changing settings in the midst of creation. FF you just plug and chug - ROpe, you can tweak the settings back and forth so that you can perfect what needs to be done while the video is being created 3) The settings in Rope are advanced and you can do much more than what FF offers - FF again is basically just plug it in and go. 4) FF was literally lifted from Rope. The developer essentially used the Rope code to make his own .. now thats fine because its open-source, but there is no actual ingenuity. It's Rope lite with a little bit nicer looking GUI Personal: I actually dont support FF because the developer has proven himself to be a bad faith actor. There was a big ordeal where I had to call him out to the open-source community because he was unjustly intimidating and threatening open-source developers (it became a big thing on Reddit and he was falsely using the DMCA as a weapon). As well, there is litany of documented testimonies of users that he abuses and belittles. I've had the unfortunate opportunity of interacting with him and he's truly an egregious man and not someone I would support. So.. on a technical level, ROPE is the best software there is anyways. On a personal level, I wouldn't use his software even if it was decent... which in comparison with Rope, it is not.
@DannySmith-bg1pk 2 күн бұрын
@@cognibuild Hey, thanks for explaining that! I didn't know FF was forked from Rope. No wonder it felt a bit... derivative. I always found myself missing features Rope had. You cleared things up for me.
@Alex_Niko_Y 3 күн бұрын
A comment in support of a better channel! FluxGym is a great program for training Lora Flux, I tried One Trainer, but it is more complicated in settings and I did not like the result, FluxGym is cooler👍
@cool-hand-luke627 3 күн бұрын
Get going fast two thumbs up👍👍. Dead simple Hunyuan video lora training tutorial for two big Hugs!
@Elwaves2925 3 күн бұрын
My main issue with FluxGym, which isn't the fault of the software, is that using 20-30 images takes way too long on 3060 with 12Gb VRAM - a full day. For that amount it's much easier and quicker to use a paid service. However, I recently tried 6 images at 1024x for 6 epochs and 10 repeats and it still produced fantastic results in 3 hours. Much better if you only have a few images available of the subject.
@MikhailBY 3 күн бұрын
Thanks for the video a lot! For the everyone who has the following error message: site-packages\safetensors\torch.py", line 313, in load_file with safe_open(filename, framework="pt", device=device) as f: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ FileNotFoundError: No such file or directory: ".\\weights\\model.safetensors" Or --weights_dir is required Fixed by resetting OmniParser master to 8a3c07d0cdb726d51ca513c446316b28118364d1, comment: update eadme, safetensor
@gamersgabangest3179 3 күн бұрын
I have tried all of them. FluxGym is the best when it works. computer constantly crashes, even though I have a 4070 superTi 16GB. The best so far, not the easiest, is Kohya with Block swipe enabled. It's faster than FluxGym and Loras are pretty big (170Mb vs 18Mb of FluxGym). FluxGym was developed by the same person who made Pinokio and he is not longer updating it.
@giovannioro1642 4 күн бұрын
Kijai has updated Hunyuan witb TeaCache!
@cognibuild 4 күн бұрын
awesome! thanks for the heads up
@smithff8381 4 күн бұрын
I'm from Brazil, I really liked your video, do you know how to solve the error Error The size of tensor a (640) must match the size of tensor b (320) in non-singleton dimension 1
@cognibuild 4 күн бұрын
i think.. try different size image
@vitaliyklitchko6664 4 күн бұрын
Thank you for the useful content. In regards the face restorer: you could just reduce blend level for the Restorer... Also there is noticeable box around face.
@cognibuild 4 күн бұрын
@@vitaliyklitchko6664 thank you for the tip with the restore. Yes that box around the face is something that if I'm spending more time I will blend the colors and all that. For this tutorial though I was just doing a quick things. Have you found a good solution for that?
@Seagull317 4 күн бұрын
Can I keep running this f5-tts in offline mode? Even after restarting the computer?
@cognibuild 4 күн бұрын
@@Seagull317 yes absolutely. It's installed on your computer and so as long as you're not downloading KZbin video you're fine
@luthfan9395 4 күн бұрын
bro why you did not put subtitle english auto generated in this video because i have language barrier, please put subtitle english auto generated in this video
@davidchamberlain6466 5 күн бұрын
Thanks. I followed the relevant parts of your tutorial (I have comfyui installed with Stability Matrix so some of the files go into different places) and I finally have something working in ComfyUI which, until now, has scared the sh1t out of me. First try was a dancing granny and it is pretty impressive even at a low resolution (480x480).
@cognibuild 5 күн бұрын
@@davidchamberlain6466 that's great! Good job
@THENOITALL 5 күн бұрын
So i need to install Vs Build Tools, Cuda, CuDNN, Fresh AI?, Miniconda, Python, Git and WSL to get LLamafactory running?
@cognibuild 5 күн бұрын
@@THENOITALL you need to have those things installed to run any AI. It's an issue of having your system set up correctly. I have an installer script that downloads them all if you require assistance
@alberttom1610 5 күн бұрын
Hi, help please, I install exactly same version found on the Website Old archive of downloads. I install right click as administrator. only user, uncheck.. and still getting "conda is not recognized as an internal or external command" what am I doing wrong. I'm on step 3 of installing MagicQuill.
@muratgelis3109 5 күн бұрын
please dont laugh at me :)) i manage to installation open it and after close i can not open because there is no shortcut on the menu..how can i run the progremme again? thank you so much btw appreciate it great effort.
@cognibuild 5 күн бұрын
@@muratgelis3109 go into your installation folder and look for a start or a main.py file
@zoey02024 5 күн бұрын
when I start the f5-tts_infer-gradio I get the following error: E:\miniconda3\envs\f5-tts\lib\site-packages\gradio\components\chatbot.py:242: UserWarning: You have not specified a value for the `type` parameter. Defaulting to the 'tuples' format for chatbot messages, but this is deprecated and will be removed in a future version of Gradio. Please set type='messages' instead, which uses openai-style dictionaries with 'role' and 'content' keys. warnings.warn( Starting app... 信息: 用提供的模式无法找到文件。 It doesn't work.How do I fix this?
@cognibuild 5 күн бұрын
@@zoey02024 try updating your gradio
@zoey02024 5 күн бұрын
@@cognibuild I tried to update gradio with this command pip install --upgrade gradio, but the problem still exists, according to the tip I should add type=‘messages’ to the chatbot.py file, but I don't know how to change it and where to add it to the file.
@JoshuaKoonce-ex5in 6 күн бұрын
I don't have a folder named python_embeded. So what do I do? ComfyUI works great.
@cognibuild 6 күн бұрын
@@JoshuaKoonce-ex5in that means you don't have the portable version which is comfy which is fine. So use the virtual environment you created in the comfy folder
@lzafar 6 күн бұрын
Bro needs to sort out his 1980’s level of green screen ability before giving news of cutting edge visual arts 😂
@comiccultivation 6 күн бұрын
Excellent vid, thanks
@captainadventures 6 күн бұрын
Excellent
@JohnnyDetour 7 күн бұрын
I tried the local version install following your excellent tutorial but failed at end as I have INTEL Graphics not NVIDIA. so before I attempt this, will this be an issue also?
@cognibuild 7 күн бұрын
@@JohnnyDetour I think that this tutorial should be fine. The main thing is that you want to be looking for if they're having you install stuff with cuda... When you see that the instructions require cuda than you're probably out of luck. However these instructions seem to work. You can also try the comfy UI installation github.com/magic-quill/MagicQuill
@aivideos322 7 күн бұрын
Glad you like the workflow, thanks for the video and praise :) i have more to add but i am worried it will get too complex for many. I have many workflows that get crazy complex and i have a tendency to complicate things for many. Nice to have someone explain things as i can only do so much with notes. Audio end time does not need to be set. Just the start seconds are needed, it will auto cut the audio to the video length.
@Alex_Niko_Y 7 күн бұрын
Thanks for the video and interesting workflow! Happy New Year 2025!!!
@propelproject 8 күн бұрын
thanks! how do you join videos together for a longer scene then 6sec?
@hannibaltezozomoctli587 8 күн бұрын
Hey Gary, do you think you could do a tutorial for Fish Speech TTS? They just released a 1.5 model and it is quite impressive, especially the emotional fluctuations in the voice. Your tutorial on F5 TTS was amazing and very straight forward. But I can't install Fish Speech correctly and it would be amazing if you could do a similar tutorial for that one. I am not computer savvy so when I tried to follow the command prompt instructions on the github page I was just facing one issue after the next. Im sure I am just making a dumb mistake but I can't figure it out. You are the only KZbinr that broke down and explained miniconda to me clearly btw.
@cognibuild 7 күн бұрын
@@hannibaltezozomoctli587 sure I'll look into it. I hadn't heard of that. It'll be a couple days as I'm out of town but I'll start looking into it today
@propelproject 8 күн бұрын
Thank you, bro! My dishes are actually going to be done now, or not. How do I add a lora into the work flow so the video has a specific character?
@harmonyreform 8 күн бұрын
"If they did not write this in here ... the Chinese government will take this down!" Then a big surprise after! 🤣🤣🤣 Come on bro, Chinese people are not living in the cave, many Chinese people are aware of the criticisms that some Westerners have about China, but most of these critiques are seen as either untrue or not reflective of the Chinese perspective. This disconnect in understanding is why it’s so difficult for some Westerners to truly grasp the complexities of China and its people. You should really listen to the perspectives of actual Chinese people rather than relying solely on media portrayals
@cognibuild 8 күн бұрын
@@harmonyreform you need to understand how communist regimes work. the Chinese people are great but the government is oppressive and totalitarian regardless of someone's view of it. All you need to do is look in the news to see how the Chinese government handles propaganda. That's not even a political statement I'm making just a factual oneAnyway the model did do those things the first times I played with them.
@harmonyreform 8 күн бұрын
@@cognibuild The comment wasn’t directed at you personally-it was more about using your video as a way to encourage some people in the West to see things from a different perspective. Thanks for your work!
@jonathaningram8157 8 күн бұрын
for some reason this model keeps making still image and adding subtitles.
@cognibuild 8 күн бұрын
@@jonathaningram8157 some videos do that. And depending on the workflow you just need to change the settings. Also try changing the seed
@sarpsomer 8 күн бұрын
Good fashion choice!
@DM-dy6vn 8 күн бұрын
26:05 She is driving backwards 😁
@Baka_Oppai 8 күн бұрын
I was putting off ltx because i thought hunyuan was gonna be better. do i have to go back now and do this?
@cognibuild 8 күн бұрын
@@Baka_Oppai I would. It's definitely fulfilling and fun to play with.. but I would only do it with image to video
@AdnanNuh 8 күн бұрын
Look Mr. Trump has no cowboy hat
@dedmaroz8943 8 күн бұрын
@21:53 the main reason it's not looking right is that you have your Occluder slider set to +24. Majority of the time, the slider for the Occluder and/or the DFL XSeg (which is the occluder taken from Deep Face Lab) should be set between -3 and +3. I mostly use the DFL XSeg, but either of these will help with things like hair on the forehead, glasses, sideburns, by giving you control over whether they are shown in the final result or not. Great deep dive into this excellent software, thank you
@Zyran.n 8 күн бұрын
Can i run this on rtx 2060s 8gb vram
@cognibuild 8 күн бұрын
@@Zyran.n I released a video on how to run on 8gn that you can try
@cognibuild 8 күн бұрын
Within the hour after posting this video version 3 was released! Be sure to download the new version
@petergreene8760 8 күн бұрын
U know what's funny. After u drop this video. 1 hour later. ROCKY533 released a new Version "V3". I was confused why my workflow didn't match urs. Then I looked, u were using V2 and I was using V3. Lol
@cognibuild 8 күн бұрын
@@petergreene8760 lol that's impeccable timing
@blade6568 9 күн бұрын
Your microphone picked it well
@cognibuild 9 күн бұрын
NOTE: Correction to workflow. When using later: Click "Switch to Own Prompt" = True