Stable Diffusion - Checkpoints and LoRAs the Basics - Fooocus
Жүктеу.....
Пікірлер: 67
@KLEEBZTECH4 ай бұрын
If you find the video useful and would like to tip, KZread does have the Super Thanks option. Those are greatly appreciated so I can have my coffee!! ☕ BUY ME A COFFEE: ko-fi.com/kleebztech
@mbcase7 ай бұрын
Thanks for an honest and complete description of how to do this process. Very refreshing compared to the usual clickbait about Midjourney and other AI image/video processes.
@KLEEBZTECH
7 ай бұрын
You're welcome and thank you very much for the compliment. I am my own worst critic so these comments help me keep doing it. I do plan on doing videos on other things but I won't trick anybody into anything. Honestly I try to be as clear as possible because I don't want anybody clicking on it that is not going to watch much of it. KZread penalizes for low retention.
@jonasopina352925 күн бұрын
wow! I have been learning a lot with your videos and examples! you just got a new subscriber! keep making great videos!!
@KLEEBZTECH
25 күн бұрын
Welcome aboard! I have a bunch and will continue as long as I can.
@user-qg8qc5qb9r27 күн бұрын
Introduction - 00:00:01 Saving Checkpoints and LoRAs - 00:00:27 Downloading and Saving Models - 00:01:00 Where to Get Checkpoints and LoRAs - 00:03:15 Downloading Checkpoints from Civit AI - 00:03:56 Changing Checkpoints in Fooocus - 00:06:15 Using Refiners in Fooocus - 00:07:26 Introduction to LoRAs - 00:08:34 Downloading and Using LoRAs - 00:09:46 Testing Effects of LoRAs - 00:10:58 Example: Adjusting LoRA Weight - 00:11:58 Combining Checkpoints and LoRAs - 00:14:33 Conclusion and Upcoming Videos - 00:15:10
@Bamigaru6 ай бұрын
Hello! I wanted to take a moment to express my gratitude for your tutorial series. The commitment and dedication you put into creating high-quality content are evident. Your ability to explain complex concepts clearly, including a multitude of practical examples, is truly exceptional. I deeply appreciate that you take nothing for granted and show every step of the process without cuts. This detail, so rare to find, makes your work even more valuable. It’s refreshing to see content that breaks away from the norm of the usual clickbait videos, which unfortunately seem to dominate this topic, while your authentic and honest approach. I try to comment as much as possible, hoping to contribute to bringing more visibility to your videos on the KZread homepage. As my way of thanking you for the high-quality videos you produce. Keep it up! I'll stop here with the videos for today, to give myself time to digest all the information you've provided. i cant wait to continuing the series tomorrow :) Love from Italy!
@KLEEBZTECH
6 ай бұрын
Thank you for the feedback. I always wonder if my approach is what people like. I know many other videos cover the same subjects in a fraction of the time which I understand helps in some ways with viewer retention which looks better to KZread. I sometimes worry I ramble on to much. lol. As for cuts.... There are many but I try not to cut out anything important. Just me rambling on too much. 😂
@vspolley7 ай бұрын
Thanks again! I've been enjoying playing around with Fooocus.
@KLEEBZTECH
7 ай бұрын
My pleasure! I plan on making many more and diving deeper. The biggest thing to remember is to keep experimenting.
@heinzwillikoch2715
7 ай бұрын
Thanks
@Miss_Marsi5 ай бұрын
This is exactly what I was looking for! Thanks so much for making this vid!
@KLEEBZTECH
5 ай бұрын
Glad it was helpful!
@landyandy6 ай бұрын
Very informative. Thank you.
@KLEEBZTECH
6 ай бұрын
You are welcome!
@GES19853 ай бұрын
Instead of being able to generate 4 pics from one prompt on one model, it would be neat to select 4 models and generate 1 pic from each off of 1 prompt. Would make comparing models easier
@KLEEBZTECH
3 ай бұрын
The only issue with that would be the time it would take to switch models while doing that. It would have to load each model before generating.
@asheshsrivastava89216 ай бұрын
Great video. Hey can you make a video showing how to train lora?
@KLEEBZTECH
6 ай бұрын
I would love to. But my past experiences doing it locally have not gone well. My system really can't handle it. I have not tried recently but plan on looking into it again and see if things have changed. If so I will probably make a video on it. 👍
@asheshsrivastava8921
6 ай бұрын
@@KLEEBZTECH I use rundiffusion a lot.
@KLEEBZTECH
6 ай бұрын
I haven't tried them out. Currently I've been trying to do mostly local as much as possible. Would you say they're pretty good? Reasonably priced?
@asheshsrivastava8921
6 ай бұрын
@@KLEEBZTECH they are pretty good. Pricing is a bit expensive.
@armondtanz
5 ай бұрын
@@KLEEBZTECH Damn was hoping to find out out about training loras :(
@petermclennan67816 ай бұрын
Excellent tutorials. Your audio levels could be a little higher.
@KLEEBZTECH
6 ай бұрын
Thank you. Yeah I'm still trying to get the audio right. I'm using a different recording setup than I used to use in the past.
@jeffreyblissi62916 ай бұрын
Hello beautiful souls. I like to ask you about my situation. I don't have a nvidia grafic card in my computer and the fooocus is give me the sign error. I do run fooocus online with the cloud version.. Can I add checkpoints and loras in the cloud version? Blessings
@musicacuantica
6 ай бұрын
I have the same question!
@nibiruenki12755 ай бұрын
hi, I'm Italian, and I follow your tutorials on "fooocus" which are the best on the net, I have this problem. How can I make it read models other than SDXL that are not supported? Here is the error fooocus displays to me: (You have selected base model other than SDXL. This is not supported yet.) Thanks for your help😄
@KLEEBZTECH
5 ай бұрын
If you are trying to use models like SD 1.5 you need to use an SDXL based checkpoint as the primary and then can use a 1.5 as the refiner and then also can use 1.5 LoRAs. When you select a refiner model it will give you suggestions on when to have the refiner switch in the small print under the model selections. Hopefully that gets you in the right direction. I do not have a video on that yet.
@nibiruenki1275
5 ай бұрын
@@KLEEBZTECH Number One!!! everything works perfectly. Thank you very much, I've been racking my brain all afternoon.
@ievajonynaite952723 күн бұрын
How did you train your own LoRA? Can you do a video on that?
@KLEEBZTECH
23 күн бұрын
I used an online service that no longer exists. But I am now experimenting with CivitAI and their LoRA training. You can earn buzz (their credits) for sharing images and that can then be used to train LoRAs. I will probably be doing a video soon on that once I get a better understanding of how it works and I get my first good one done.
@ievajonynaite9527
23 күн бұрын
@KLEEBZTECH amazing, can’t wait! Thanks
@osmundmillion32994 ай бұрын
Any recommendations for good LoRas bro? Anime and Realistic.
@MrJanchoy19805 ай бұрын
Thx @KLEEBZTECH , I reall need this toturial😍
@KLEEBZTECH
5 ай бұрын
I do recommend some of my other videos as well. 😉 kzread.info/head/PLSofK8pvSk5oHH9iNAAemQREADEY7CPp_
@TheTchaume6 ай бұрын
Thanks for all thoses great videos aboutfooocus. Do you know if it's possible to directly generate images with higer resolutions ? And do you know it's possible to save the basic settings when you start fooocus (resolution, quality, loras etc). Thanks !
@KLEEBZTECH
6 ай бұрын
The simple answer to your first question is no. As for saving the settings you can edit the config file so the defaults are what you want. kzread.info/dash/bejne/nHuE2tiLf6_dhbg.html
@treapymnm65195 ай бұрын
hI bro . for some reason it only work with the base models for me (namely that every sdxl i downlaod doesnt load ) how can i fix it?
@KLEEBZTECH
5 ай бұрын
First question is if you are using locally? And if you could explain a little more detail what you mean by it only working with the base models.
@VigneshoViyan7 ай бұрын
how to add custom path
@KLEEBZTECH
7 ай бұрын
If you want to change any of the locations for models or output there is a config.txt file in the Fooocus folder that you can edit.
@musicalstation1017 ай бұрын
Can u please do how they create those ai influencer with those tools where videos and pictures look very real and unique mixing different ai
@KLEEBZTECH
7 ай бұрын
I do plan on covering other tools in future videos. You can get very real looking stuff from Fooocus with some practice. There are so many different tools coming out it is difficult to keep up but I am trying. lol.
@lilillllii2464 ай бұрын
Thank you. It's a little different question, but how can I use the eular a sampler in fooocus?
@KLEEBZTECH
4 ай бұрын
You can change those in the debug menu.
@lilillllii246
4 ай бұрын
@@KLEEBZTECH there is no eural a in sampler of debg
@KLEEBZTECH
4 ай бұрын
Sorry about that. Not on computer and didn't realize that was not one of the available samplers in the list. @@lilillllii246
@KLEEBZTECH
4 ай бұрын
I just got on my computer and can confirm "euler_ancestral" should be listed in the sampler drop down menu.
@KLEEBZTECH
4 ай бұрын
It should be the second one on the list if you scroll to the top of the list.
@Hobbies-and-Experiments6 ай бұрын
Nope, this is bad, when I go to that civit site i hear about 10 voices all talking at the same time and no way to stop it. I see nothing like you have there when it comes down to the model download, all the tabs are different and nothing makes sense. but thank you for what you did provide as now i can go to huggingface and get some models, so thanks 👍
@KLEEBZTECH
6 ай бұрын
Are you going to the link in the description of the video? Because what you describe does not sound like the same site. Huggingface is a good source but much harder to find what you are looking for which is why it is not as popular. civitai.com/
@Hobbies-and-Experiments
6 ай бұрын
@@KLEEBZTECH ya I think I might have ended up on a scam site lol. all good now and I see it the same as yours, thanks buddy for all the info I appreciate it.
@TheHistoryCode1256 ай бұрын
I only use the gradio not local
@cesarluisdupuy12935 ай бұрын
can this be done in google colab?
@KLEEBZTECH
5 ай бұрын
I do know that you can change them in colab but I have not used Fooocus there myself so not sure of how it is done. Might be some info on the Github discussions for Fooocus. Lots of information there.
@LahiruBandara-iq8xd2 ай бұрын
How to do this in Google colab
@KLEEBZTECH
2 ай бұрын
I have one on Colab. Although I have now changed how I do things a little. I use my Google Drive for models and share a link to use in Colab so it is almost instant to grab them. You can also upload LoRA into the folder in Colab. kzread.info/dash/bejne/YmuZztWFgbicmNI.html
@phalxor4 ай бұрын
you dont need a 1.5 refiner to use 1.5 models
@KLEEBZTECH
4 ай бұрын
You need a 1.5 model as a refiner to use 1.5 Loras.
@tomasm12333 ай бұрын
When training Lora of your goat, did you use captions .txt file or blips? I am looking to do first my ever training of the face thought. Likeness of the person face is very important to me, though.
@KLEEBZTECH
3 ай бұрын
Honestly I used an online service that took care of most of the process. That service doesn't exist anymore. I've been messing around with training them myself as well now but haven't perfected it yet. I'm using text files and I have it automatically generate them and then edit them after so they are better. I find automated captioning isn't the greatest. But I'm no expert on doing those myself.
@klb1113
Ай бұрын
@@KLEEBZTECH I keep hearing about this software called Kohya that can run locally, so I'd love to eventually see your take on it. (Once you have a computer that can handle it, that is!)
@KLEEBZTECH
Ай бұрын
@@klb1113 I do plan on giving it a shot locally soon. I am hoping that is has gotten better and that I can now run it without much issues. I have been playing with OneTrainer as well. Problem is that it takes so long for these on my system that testing is very time consuming.
Пікірлер: 67
If you find the video useful and would like to tip, KZread does have the Super Thanks option. Those are greatly appreciated so I can have my coffee!! ☕ BUY ME A COFFEE: ko-fi.com/kleebztech
Thanks for an honest and complete description of how to do this process. Very refreshing compared to the usual clickbait about Midjourney and other AI image/video processes.
@KLEEBZTECH
7 ай бұрын
You're welcome and thank you very much for the compliment. I am my own worst critic so these comments help me keep doing it. I do plan on doing videos on other things but I won't trick anybody into anything. Honestly I try to be as clear as possible because I don't want anybody clicking on it that is not going to watch much of it. KZread penalizes for low retention.
wow! I have been learning a lot with your videos and examples! you just got a new subscriber! keep making great videos!!
@KLEEBZTECH
25 күн бұрын
Welcome aboard! I have a bunch and will continue as long as I can.
Introduction - 00:00:01 Saving Checkpoints and LoRAs - 00:00:27 Downloading and Saving Models - 00:01:00 Where to Get Checkpoints and LoRAs - 00:03:15 Downloading Checkpoints from Civit AI - 00:03:56 Changing Checkpoints in Fooocus - 00:06:15 Using Refiners in Fooocus - 00:07:26 Introduction to LoRAs - 00:08:34 Downloading and Using LoRAs - 00:09:46 Testing Effects of LoRAs - 00:10:58 Example: Adjusting LoRA Weight - 00:11:58 Combining Checkpoints and LoRAs - 00:14:33 Conclusion and Upcoming Videos - 00:15:10
Hello! I wanted to take a moment to express my gratitude for your tutorial series. The commitment and dedication you put into creating high-quality content are evident. Your ability to explain complex concepts clearly, including a multitude of practical examples, is truly exceptional. I deeply appreciate that you take nothing for granted and show every step of the process without cuts. This detail, so rare to find, makes your work even more valuable. It’s refreshing to see content that breaks away from the norm of the usual clickbait videos, which unfortunately seem to dominate this topic, while your authentic and honest approach. I try to comment as much as possible, hoping to contribute to bringing more visibility to your videos on the KZread homepage. As my way of thanking you for the high-quality videos you produce. Keep it up! I'll stop here with the videos for today, to give myself time to digest all the information you've provided. i cant wait to continuing the series tomorrow :) Love from Italy!
@KLEEBZTECH
6 ай бұрын
Thank you for the feedback. I always wonder if my approach is what people like. I know many other videos cover the same subjects in a fraction of the time which I understand helps in some ways with viewer retention which looks better to KZread. I sometimes worry I ramble on to much. lol. As for cuts.... There are many but I try not to cut out anything important. Just me rambling on too much. 😂
Thanks again! I've been enjoying playing around with Fooocus.
@KLEEBZTECH
7 ай бұрын
My pleasure! I plan on making many more and diving deeper. The biggest thing to remember is to keep experimenting.
@heinzwillikoch2715
7 ай бұрын
Thanks
This is exactly what I was looking for! Thanks so much for making this vid!
@KLEEBZTECH
5 ай бұрын
Glad it was helpful!
Very informative. Thank you.
@KLEEBZTECH
6 ай бұрын
You are welcome!
Instead of being able to generate 4 pics from one prompt on one model, it would be neat to select 4 models and generate 1 pic from each off of 1 prompt. Would make comparing models easier
@KLEEBZTECH
3 ай бұрын
The only issue with that would be the time it would take to switch models while doing that. It would have to load each model before generating.
Great video. Hey can you make a video showing how to train lora?
@KLEEBZTECH
6 ай бұрын
I would love to. But my past experiences doing it locally have not gone well. My system really can't handle it. I have not tried recently but plan on looking into it again and see if things have changed. If so I will probably make a video on it. 👍
@asheshsrivastava8921
6 ай бұрын
@@KLEEBZTECH I use rundiffusion a lot.
@KLEEBZTECH
6 ай бұрын
I haven't tried them out. Currently I've been trying to do mostly local as much as possible. Would you say they're pretty good? Reasonably priced?
@asheshsrivastava8921
6 ай бұрын
@@KLEEBZTECH they are pretty good. Pricing is a bit expensive.
@armondtanz
5 ай бұрын
@@KLEEBZTECH Damn was hoping to find out out about training loras :(
Excellent tutorials. Your audio levels could be a little higher.
@KLEEBZTECH
6 ай бұрын
Thank you. Yeah I'm still trying to get the audio right. I'm using a different recording setup than I used to use in the past.
Hello beautiful souls. I like to ask you about my situation. I don't have a nvidia grafic card in my computer and the fooocus is give me the sign error. I do run fooocus online with the cloud version.. Can I add checkpoints and loras in the cloud version? Blessings
@musicacuantica
6 ай бұрын
I have the same question!
hi, I'm Italian, and I follow your tutorials on "fooocus" which are the best on the net, I have this problem. How can I make it read models other than SDXL that are not supported? Here is the error fooocus displays to me: (You have selected base model other than SDXL. This is not supported yet.) Thanks for your help😄
@KLEEBZTECH
5 ай бұрын
If you are trying to use models like SD 1.5 you need to use an SDXL based checkpoint as the primary and then can use a 1.5 as the refiner and then also can use 1.5 LoRAs. When you select a refiner model it will give you suggestions on when to have the refiner switch in the small print under the model selections. Hopefully that gets you in the right direction. I do not have a video on that yet.
@nibiruenki1275
5 ай бұрын
@@KLEEBZTECH Number One!!! everything works perfectly. Thank you very much, I've been racking my brain all afternoon.
How did you train your own LoRA? Can you do a video on that?
@KLEEBZTECH
23 күн бұрын
I used an online service that no longer exists. But I am now experimenting with CivitAI and their LoRA training. You can earn buzz (their credits) for sharing images and that can then be used to train LoRAs. I will probably be doing a video soon on that once I get a better understanding of how it works and I get my first good one done.
@ievajonynaite9527
23 күн бұрын
@KLEEBZTECH amazing, can’t wait! Thanks
Any recommendations for good LoRas bro? Anime and Realistic.
Thx @KLEEBZTECH , I reall need this toturial😍
@KLEEBZTECH
5 ай бұрын
I do recommend some of my other videos as well. 😉 kzread.info/head/PLSofK8pvSk5oHH9iNAAemQREADEY7CPp_
Thanks for all thoses great videos aboutfooocus. Do you know if it's possible to directly generate images with higer resolutions ? And do you know it's possible to save the basic settings when you start fooocus (resolution, quality, loras etc). Thanks !
@KLEEBZTECH
6 ай бұрын
The simple answer to your first question is no. As for saving the settings you can edit the config file so the defaults are what you want. kzread.info/dash/bejne/nHuE2tiLf6_dhbg.html
hI bro . for some reason it only work with the base models for me (namely that every sdxl i downlaod doesnt load ) how can i fix it?
@KLEEBZTECH
5 ай бұрын
First question is if you are using locally? And if you could explain a little more detail what you mean by it only working with the base models.
how to add custom path
@KLEEBZTECH
7 ай бұрын
If you want to change any of the locations for models or output there is a config.txt file in the Fooocus folder that you can edit.
Can u please do how they create those ai influencer with those tools where videos and pictures look very real and unique mixing different ai
@KLEEBZTECH
7 ай бұрын
I do plan on covering other tools in future videos. You can get very real looking stuff from Fooocus with some practice. There are so many different tools coming out it is difficult to keep up but I am trying. lol.
Thank you. It's a little different question, but how can I use the eular a sampler in fooocus?
@KLEEBZTECH
4 ай бұрын
You can change those in the debug menu.
@lilillllii246
4 ай бұрын
@@KLEEBZTECH there is no eural a in sampler of debg
@KLEEBZTECH
4 ай бұрын
Sorry about that. Not on computer and didn't realize that was not one of the available samplers in the list. @@lilillllii246
@KLEEBZTECH
4 ай бұрын
I just got on my computer and can confirm "euler_ancestral" should be listed in the sampler drop down menu.
@KLEEBZTECH
4 ай бұрын
It should be the second one on the list if you scroll to the top of the list.
Nope, this is bad, when I go to that civit site i hear about 10 voices all talking at the same time and no way to stop it. I see nothing like you have there when it comes down to the model download, all the tabs are different and nothing makes sense. but thank you for what you did provide as now i can go to huggingface and get some models, so thanks 👍
@KLEEBZTECH
6 ай бұрын
Are you going to the link in the description of the video? Because what you describe does not sound like the same site. Huggingface is a good source but much harder to find what you are looking for which is why it is not as popular. civitai.com/
@Hobbies-and-Experiments
6 ай бұрын
@@KLEEBZTECH ya I think I might have ended up on a scam site lol. all good now and I see it the same as yours, thanks buddy for all the info I appreciate it.
I only use the gradio not local
can this be done in google colab?
@KLEEBZTECH
5 ай бұрын
I do know that you can change them in colab but I have not used Fooocus there myself so not sure of how it is done. Might be some info on the Github discussions for Fooocus. Lots of information there.
How to do this in Google colab
@KLEEBZTECH
2 ай бұрын
I have one on Colab. Although I have now changed how I do things a little. I use my Google Drive for models and share a link to use in Colab so it is almost instant to grab them. You can also upload LoRA into the folder in Colab. kzread.info/dash/bejne/YmuZztWFgbicmNI.html
you dont need a 1.5 refiner to use 1.5 models
@KLEEBZTECH
4 ай бұрын
You need a 1.5 model as a refiner to use 1.5 Loras.
When training Lora of your goat, did you use captions .txt file or blips? I am looking to do first my ever training of the face thought. Likeness of the person face is very important to me, though.
@KLEEBZTECH
3 ай бұрын
Honestly I used an online service that took care of most of the process. That service doesn't exist anymore. I've been messing around with training them myself as well now but haven't perfected it yet. I'm using text files and I have it automatically generate them and then edit them after so they are better. I find automated captioning isn't the greatest. But I'm no expert on doing those myself.
@klb1113
Ай бұрын
@@KLEEBZTECH I keep hearing about this software called Kohya that can run locally, so I'd love to eventually see your take on it. (Once you have a computer that can handle it, that is!)
@KLEEBZTECH
Ай бұрын
@@klb1113 I do plan on giving it a shot locally soon. I am hoping that is has gotten better and that I can now run it without much issues. I have been playing with OneTrainer as well. Problem is that it takes so long for these on my system that testing is very time consuming.