Hi. If you are meaning the general 'bluriness' of the deepfake itself, that could be one of a few things. First off, it could be the resolution of the deepfake itself. Like a still-image or video resolution, the higher this is...the sharper the final result "can" be. So if you have a GPU with a lot of VRAM, then you can use a higher quality setting. For my own I have to keep the quality in the middle; 8GB is the VRAM on my card. Second, it could be how long you train for. Generally, the longer you give it...the cleaner the results will be. There gets to a point of diminishing returns, but there are times where I thought it was "good enough", let it go longer, and did see an improvement. Lastly, the GAN step outlined in the third video in this series can help bring out more detail and sharpen things up. Use it sparingly though -- it can cause artifacting if put too high. Hope some of this helps! :)
Probably the only understandable tutorial about this matter. Thanks Johnny. Do you know how long should I let my gtx 1070 work per day without risking damage by overheating? Or is it a myth? I see it´s running between 69 and 74 °C by now.
Hi! Great question. I'm often training / running on a 1080 FE, so not hugely different than your 1070. That temperature is likely fine, from what is coming to mind. I've had a multi-gpu system with "blower style" cards packed in, and the card in the middle got much higher in temperature than the other two. So in that scenario I think temperature was a concern -- but a well-ventilated case at 69-74 I think should be fine; even if it's a few days, I don't think it should damage. Obviously I can't guarantee. I'd typically leave the cards running at night (would get hot to have them running when I was in the room for long periods of time) or if I was using the PC & still wanted it to continue , I'd keep one of the other GPU's running while the primary (2080) was being used for editing or regular PC usage, etc. Never really had any problems in these scenarios, so I hope this helps a bit! :) EDIT: Just a FYI...We don't do deepfakes all the time, so the cards aren't getting massive amounts of work put on them. If you're doing deepfakes 24x7, then that might end up being a different story...then you'd be doing similar things to bitcoin mining which (supposedly) can hurt a card, but even that's debated. As long as the room that the PC is in doesn't get too hot though & the case / GPU is ventilating well enough though, I think you'd be fine for 'regular usage'.
I'm having this error I try all my best to solve it but was still unable failed to load the native TensorFlow runtime Please help me a good on StackOverflow about it but still, I'm unable to solve it
i cant train from the gpu. it always trains with the cpu and this takes very long. i have the geforce gtx 650 4gb. Any idea how i can train with the gpu?
Hmm, this is interesting. I know that the newer version require an Nvidia GPU (which obviously you have), but I'm not sure what the overall requirements are beyond that. I would definitely try updating drivers if you haven't already, but from there you might have to dig a bit and see if your model of GPU is supported. The 650 is still a plenty good card, but I'm not sure where the cutoff is in the Nvidia lineage for Deepfacelab. If you figure things out, please let me know; curious to hear! :)
@@JhowT I already updated the newest driver of my graphic card but th same issue appears. i‘ve heard that the tensorflow version of the DFL may be doesn‘t support the gtx 650. Im not sure what to do. Thnx for answering my question.
@@montajabi Yeah I'm sorry I don't have more information for you. There are definitely different branches at this point, with even a newer one appearing right after the new year only for Nvidia 30-series cards. I suppose your best course of action would be to work backwards in branches until you find the newest branch that works with your card? I've never used v1.0 of the software, but if v2 doesn't work (which I think is all tensor flow) maybe try v1 ?
Hmm, that's a good question. You might have to dig into the technical requirements for the software to see, or it might come down to the VRAM you have on the card...since that will really determine how large of a [usable] resolution you can render out.
Heck yes, and beyond that...there is a newer version / branch of the software that's made specifically for 30-series Nvidia cards, which I'm assuming are accelerated beyond what most cards are capable of.
@@JhowT wow really my friend ? so i guess that s prety fine haha where to find this version ? on GitHub ? can you share me the link please i didnt find this version, thanks you
@@CristianoRonaldo-zy3pq Hi, this is the link...I think it's public for everyone to access. You would grab the one that's NOT titled "for previous", since you've got the latest tech. ;)
Hi! Most of the links are in the video description, which I would start with "DeepFaceLab Github". You'll have to sign up for a Github account first if you don't already have one...which is just github.com From there you should be able to follow along and get off to the races! :)
Thank you for making this video, it was very insightful! I was wondering if you were open to private tutoring and/ or available to hire. Thank you for your time
@@JhowT I guess you can answer here too...I recently learned how to do basic deepfake...I don't have a pc so I use paperspace to make my deepfake...How can I make a video that has more than one face but I want to use only one face to deepfake and leave the other as it is?
Hi! So that should definitely be do-able. When you do the "5) data_dst faceset extract.bat" step, run the "5.1) data_dst view aligned results.bat". Once the interface is up, delete any faces that you don't want to be 'trained' for the destination. This will also only those faces that are left over. Does that make sense? :)
Hi Johnny...sorry for being annoying but I had one more question for you. Can you please let me know what's the difference between choosing while face vs face? Which is more effective? I used just face and I feel like it didn't cover much of the face which makes me wonder if I should have used wf. Please let me know! Thanks!
Thank you for the reminder! This week has been a bit of a whirlwind. If you are up for shooting me a reminder next week, I will try and make sure to finally make part 2 happen. :)
good afternoon you how to remove the blur caused deepfake creation is blurry is there any technology for this.
Hi. If you are meaning the general 'bluriness' of the deepfake itself, that could be one of a few things.
First off, it could be the resolution of the deepfake itself. Like a still-image or video resolution, the higher this is...the sharper the final result "can" be. So if you have a GPU with a lot of VRAM, then you can use a higher quality setting. For my own I have to keep the quality in the middle; 8GB is the VRAM on my card.
Second, it could be how long you train for. Generally, the longer you give it...the cleaner the results will be. There gets to a point of diminishing returns, but there are times where I thought it was "good enough", let it go longer, and did see an improvement.
Lastly, the GAN step outlined in the third video in this series can help bring out more detail and sharpen things up. Use it sparingly though -- it can cause artifacting if put too high.
Hope some of this helps! :)
thanks friend and I would like to put audio in video like a video with the voice to dub the other in 2 video would put the voice.
Probably the only understandable tutorial about this matter. Thanks Johnny. Do you know how long should I let my gtx 1070 work per day without risking damage by overheating? Or is it a myth? I see it´s running between 69 and 74 °C by now.
Hi! Great question. I'm often training / running on a 1080 FE, so not hugely different than your 1070. That temperature is likely fine, from what is coming to mind.
I've had a multi-gpu system with "blower style" cards packed in, and the card in the middle got much higher in temperature than the other two. So in that scenario I think temperature was a concern -- but a well-ventilated case at 69-74 I think should be fine; even if it's a few days, I don't think it should damage. Obviously I can't guarantee. I'd typically leave the cards running at night (would get hot to have them running when I was in the room for long periods of time) or if I was using the PC & still wanted it to continue , I'd keep one of the other GPU's running while the primary (2080) was being used for editing or regular PC usage, etc. Never really had any problems in these scenarios, so I hope this helps a bit! :)
EDIT: Just a FYI...We don't do deepfakes all the time, so the cards aren't getting massive amounts of work put on them. If you're doing deepfakes 24x7, then that might end up being a different story...then you'd be doing similar things to bitcoin mining which (supposedly) can hurt a card, but even that's debated.
As long as the room that the PC is in doesn't get too hot though & the case / GPU is ventilating well enough though, I think you'd be fine for 'regular usage'.
I'm having this error I try all my best to solve it but was still unable
failed to load the native TensorFlow runtime
Please help me a good on StackOverflow about it but still, I'm unable to solve it
Are you using an Nvidia GPU that is compatible, by chance?
Awesome tutorial, very simple, concise and fast ! Thank you !
Very glad it was useful to you. :)
i cant train from the gpu. it always trains with the cpu and this takes
very long. i have the geforce gtx 650 4gb. Any idea how i can train with
the gpu?
Hmm, this is interesting. I know that the newer version require an Nvidia GPU (which obviously you have), but I'm not sure what the overall requirements are beyond that. I would definitely try updating drivers if you haven't already, but from there you might have to dig a bit and see if your model of GPU is supported. The 650 is still a plenty good card, but I'm not sure where the cutoff is in the Nvidia lineage for Deepfacelab. If you figure things out, please let me know; curious to hear! :)
@@JhowT
I already updated the newest driver of my graphic card but th same issue appears.
i‘ve heard that the tensorflow version of the DFL may be doesn‘t support the gtx 650. Im not sure what to do. Thnx for answering my question.
@@montajabi Yeah I'm sorry I don't have more information for you. There are definitely different branches at this point, with even a newer one appearing right after the new year only for Nvidia 30-series cards. I suppose your best course of action would be to work backwards in branches until you find the newest branch that works with your card? I've never used v1.0 of the software, but if v2 doesn't work (which I think is all tensor flow) maybe try v1 ?
Work in RX 580 8G?
Hmm, that's a good question. You might have to dig into the technical requirements for the software to see, or it might come down to the VRAM you have on the card...since that will really determine how large of a [usable] resolution you can render out.
is my 3080 is ok for deep fake ?
Heck yes, and beyond that...there is a newer version / branch of the software that's made specifically for 30-series Nvidia cards, which I'm assuming are accelerated beyond what most cards are capable of.
@@JhowT wow really my friend ? so i guess that s prety fine haha
where to find this version ? on GitHub ? can you share me the link please i didnt find this version,
thanks you
@@JhowT btw i really love your work ! you are amazing bro ! +1 sub
@@CristianoRonaldo-zy3pq Hi, this is the link...I think it's public for everyone to access. You would grab the one that's NOT titled "for previous", since you've got the latest tech. ;)
@@CristianoRonaldo-zy3pq Of course; you're very welcome. Glad to have you on here. :)
Really fantastic video man! :) I found this very helpful, especially the proper download of deep face lab aha lol! Thank you and God bless! :)
I'm glad it was helpful to you; appreciate you letting me know! :)
@@JhowT Thank you man! :) 👍🏾
I'm confused about how to download the software or where to get it.
Hi! Most of the links are in the video description, which I would start with "DeepFaceLab Github". You'll have to sign up for a Github account first if you don't already have one...which is just github.com From there you should be able to follow along and get off to the races! :)
Always very good explanation and great subjects
Glad you think so!
Thank you for making this video, it was very insightful!
I was wondering if you were open to private tutoring and/ or available to hire.
Thank you for your time
Hi Mai, and I'm glad to hear you found the video useful!
Feel free to ask questions on here, that way we can all learn together. :)
Nice content, wish to see more about this tutorial. Thanks for your efford !
More to come! :)
Hi! Took me a while, but wanted to let you know that Part 02 is up!
@@JhowT Thank you very much ^^
Hi there. Do you have an email address or contact info I can message you with some questions about deepfake?
Hi there. Are they general workflow questions? If so, I can try and answer on here?
@@JhowT I guess you can answer here too...I recently learned how to do basic deepfake...I don't have a pc so I use paperspace to make my deepfake...How can I make a video that has more than one face but I want to use only one face to deepfake and leave the other as it is?
Hi! So that should definitely be do-able. When you do the "5) data_dst faceset extract.bat" step, run the "5.1) data_dst view aligned results.bat". Once the interface is up, delete any faces that you don't want to be 'trained' for the destination. This will also only those faces that are left over. Does that make sense? :)
@@JhowT totally makes sense... thanks so much!
Hi Johnny...sorry for being annoying but I had one more question for you. Can you please let me know what's the difference between choosing while face vs face? Which is more effective? I used just face and I feel like it didn't cover much of the face which makes me wonder if I should have used wf. Please let me know! Thanks!
Part 2!!!!!
Thank you for the reminder! This week has been a bit of a whirlwind. If you are up for shooting me a reminder next week, I will try and make sure to finally make part 2 happen. :)
Part 02 is up! Sorry for the wait. ;)