💖💖I used this technique to create almost my entire channel using only 2Gb VRam on a GTX960. 😃With the right settings FOOOCUS allows you to do almost anything. Now I also use FORGE for animations and partly COMFYUI.😢 I'm forced to use all the tricks due to the underpowered hardware, but with the right tricks you can achieve almost anything. 💖💖 Good job thank you!
@@KLEEBZTECH 💖💖 Thanks to you for bringing so many great ideas to your channel. I'm trying to do everything with a low budget and share the techniques used to help those who don't have a 4090 to enter the world of generative AI and I see that there are excellent possibilities even for those with few resources. I'm now experimenting with using SHIFT ATTENTION in FORGE for rough animations that are doable with just the prompt. The results are encouraging... We'll see... 😊😊
I just watched your channel. It is amazing. I suggest you could also show us how you do it. specially how did you do anime style video inf Forge . thanks
@@TusharHossain 💖 Thank you for your interest in my channel content, 😊 I added some information in the replies to the other comments below regarding the settings in FOOCUS. 🤷♂ As far as Animations in FORGE are concerned, a long discussion is needed and we should ask @KLEEBZTECH 😎 if he's interested in us talking about it on his channel. 👍 However you can go to my channel and comment on the video you prefer so I can understand which styles you are referring to and I will try to create some content that explains my examples. 💖 Thank you all for the chance to talk about new exciting things.
6:28 "I'm going to go with this one here because it's safer for UA-cam, whereas a lot of the LoRAs for SD1.5 I can't really demonstrate on UA-cam." Those who don't know Those who know
You sir are a life changer. I knew how to use 1.5 Base Models with the refiner. But got confused when it came to using 1.5 Loras. You cleared this up! Thank you!
and do much more then just that with Stability Matrix have all your favorite AI gfx toys in 1 on pc wohooo im using it and its awesome. all of these steps are not really needed with Stability Matrix, you
Stability Matrix is just for managing SD UIs and resources. Yes you can use 1.5 natively with other UIs included with Stability Matrix but some people prefer Fooocus.
@@KLEEBZTECH Do you have a list of available UIs to use? Like Fooocus, Defooocus, Automatic1111, Stability Matrix, ComfyUI? Fooocus is my first UI, as I chose it because of your channel actually hahaha. I have 6gb vram and was told Fooocus was perfect for a easy setup.
If you want to experiment and use different UIs I recommend using Stability Matrix so it can manage things like shared model folders and such. You can install many different UIs in there like Invoke, Stable Swarm, A1111, ComfyUI, Fooocus for example. I use Fooocus 90% of the time myself.
Thank you very much for this video! I needed this info since the first day I began using Fooocus! I missed some of SD1.5 checkpoints. Before using Fooocus (when I was using Automatic1111) I managed to create a consistent character using a particular SD1.5 checkpoint, after migrating to Fooocus and losing that checkpoint I lost the ability to continue producing more images of that character. It was a petty. Your video gave me hope to go back to my early days with using Stable Diffusion. I can’t wait to start using SD1.5 checkpoints right now after posting this comment. Thank you very much!
Posted in pateron but thought this might be better.. Did you notice Fooocus updated and now has "playground 2.5" in the presets? I have no clue what that is and was hoping you had some info on it.
hey kleebz appreciate it cause lots of new models on 1.5 coming with better images but when i do face swap with upscale 1.5x it shows me this"Image is too large. Directly returned the SR image. Usually directly return SR image at 4K resolution yields better results than SDXL diffusion" and it returns the same img as in upscale 1.5x
@@KLEEBZTECH ok but first i mask the face and do invert mask so my face is consistent and outside of it changes in all the direction in outpaint and after that i send this image to upscale 1.5x and swap face of the img in image prompt and do img prompt and vary upscale cause i found this way better to get accurate face swap
@@Designing-hc5pz If you are outpainting that will increase the size of the image which Fooocus will only upscale to a certain size so if the image is already large it probably won't do that type of upscale.
awesome! I never knew that about the 1.5 and LoRA's! Thank you. Does it matter what Jug version I use. I see they have the X version. Think I should also make a file for the XL's too?
Well there could be LoRAs that are made for 1.5 that you could want to use and some people like 1.5 but I personally don't use it myself and would stick with SDXL personally. If you were using a UI that fully supported just 1.5 then you would be able to run it on a lower end system.
@@KLEEBZTECH V2 is a prompt word for the screen, with little change in the background. What I mean is whether there is a function similar to generating various backgrounds
How much does what cost? Fooocus is open source and does not cost anything. Be aware that there are scam sites that will try to trick you into thinking you need to pay.
Hey! Ty for idea with folders! Was manually renaming loras....)) Question here) Since I switched to 1.5 refiner seems like image prompts like pyracanny and faceswap features are not working. Am I doing smth wrong or you cant use them with 1.5 refiner?
A setting logic can be: Forced Overwrite of Sampling Step=25 or whatever Forced Overwrite of Refiner Switch Step=1 With these settings you bypass all default settings. Use SPEED mode and change everything in Developer Debug Mode including the Latent size which can be 512x512 or less for speed. Consider that you lose 1 processing step due to the Switch Step=1 so increase the Sampling Steps accordingly with the switch value and with only these parameters you configure everything to use the SD1.5 models 😉👍
@@AnimAI-LoraI’ve tried this multiple times, the results are always subpar compared to using 1.5 model in A1111. There is a slight resemblance but poor quality. Also it takes much longer to generate the image. So there is really no point.
@@WhySoBroke I would like to start by saying that FOOOCUS should be used as intended by its creator 'lllyasviel' and as expressly specified by him it is finely calibrated to work within certain parameters. 💖 For the rest you are absolutely right, 👍 because everything depends on the first Sampling Step or steps that you generate from the Base Model which leaves the Refiner with a Latent Image which will be used as a starting point.. It also depends on which Model you used as a Base both for the quality or type of result and for the speed of execution. To obtain the best possible result it is best to use two similar models such as realcartoonXL_v6 and realcartoonAnime_v10 which are from the same family but XL and 1.5. Furthermore, as Base Model you can use the smallest possible models (4-6GB) to reduce VRam requests and Pruned 1.5 models as refiners (2Gb). Set Forced Overwrite of Refiner Switch Step to 0-1. With an optimal swap setting on the SSD it is possible to generate a 512x512 image in 20 seconds using only 2GB of VRAM. For quality, consider the differences in settings between Automatic and FOOOCUS in the way they generate (ADM, CFG scale, and dozens of other parameters). In the end it's luck that decides if you've made a good setting!😉 👀 However, consider that we are talking about a fallback solution for those who do not have sufficient hardware resources and do not know how to use Automatic1111 or do not know how to configure the use of VRAM, therefore they are willing to sacrifice something in order to generate with FOOOCUS which is also easier to use.😊 In conclusion, I also use FORGE and COMFYUI with both XL and 1.5 Models, but to use them with limited hardware resources (GTX960 2Gb VrRam) you need to make appropriate settings which are not within everyone's reach. I hope it has been useful to you and that you don't take it as "the solution", because if you have done a bit of testing you will have realized that the results vary greatly even for a single comma and for each user with different configurations, hardware, software , releases and so on... 💖💖💖 A hug to everyone 😍
![Shoot EPIC MOVIES with this FREE AI Tool! [ComfyUI Tutorial + Free Workflow]](http://i.ytimg.com/vi/gHI6PjTkBF4/mqdefault.jpg)
![Lp. Сердце Вселенной #60 РОЖДЕНИЕ ЛОЛОЛОШКИ [Финал] • Майнкрафт](http://i.ytimg.com/vi/YoR0pAV9FVQ/mqdefault.jpg)
💖💖I used this technique to create almost my entire channel using only 2Gb VRam on a GTX960. 😃With the right settings FOOOCUS allows you to do almost anything. Now I also use FORGE for animations and partly COMFYUI.😢 I'm forced to use all the tricks due to the underpowered hardware, but with the right tricks you can achieve almost anything. 💖💖 Good job thank you!
Thanks for the info! Good to know.
@@KLEEBZTECH 💖💖 Thanks to you for bringing so many great ideas to your channel. I'm trying to do everything with a low budget and share the techniques used to help those who don't have a 4090 to enter the world of generative AI and I see that there are excellent possibilities even for those with few resources. I'm now experimenting with using SHIFT ATTENTION in FORGE for rough animations that are doable with just the prompt. The results are encouraging... We'll see... 😊😊
I just watched your channel. It is amazing. I suggest you could also show us how you do it. specially how did you do anime style video inf Forge . thanks
@@TusharHossain 💖 Thank you for your interest in my channel content,
😊 I added some information in the replies to the other comments below regarding the settings in FOOCUS.
🤷♂ As far as Animations in FORGE are concerned, a long discussion is needed and we should ask @KLEEBZTECH 😎 if he's interested in us talking about it on his channel.
👍 However you can go to my channel and comment on the video you prefer so I can understand which styles you are referring to and I will try to create some content that explains my examples.
💖 Thank you all for the chance to talk about new exciting things.
@AnimAI-Lora curious if you have gotten SD 1.5 embeddings to work with Fooocus?
Did not know that you could use subfolders to organize the different SD models, thanks!
6:28 "I'm going to go with this one here because it's safer for UA-cam, whereas a lot of the LoRAs for SD1.5 I can't really demonstrate on UA-cam."
Those who don't know
Those who know
You sir are a life changer. I knew how to use 1.5 Base Models with the refiner. But got confused when it came to using 1.5 Loras. You cleared this up! Thank you!
Glad to help
and do much more then just that with Stability Matrix have all your favorite AI gfx toys in 1 on pc wohooo im using it and its awesome. all of these steps are not really needed with Stability Matrix, you
Stability Matrix is just for managing SD UIs and resources. Yes you can use 1.5 natively with other UIs included with Stability Matrix but some people prefer Fooocus.
@@KLEEBZTECH Do you have a list of available UIs to use? Like Fooocus, Defooocus, Automatic1111, Stability Matrix, ComfyUI?
Fooocus is my first UI, as I chose it because of your channel actually hahaha.
I have 6gb vram and was told Fooocus was perfect for a easy setup.
If you want to experiment and use different UIs I recommend using Stability Matrix so it can manage things like shared model folders and such. You can install many different UIs in there like Invoke, Stable Swarm, A1111, ComfyUI, Fooocus for example. I use Fooocus 90% of the time myself.
Thanks!
Welcome! And thank you!
At 2:15 you can see the format in the bottom left preview link instead of clicking it ;)
I seem to forget about that on occasion...
great fooocus content, best chanel for this one!
Thank you very much for this video!
I needed this info since the first day I began using Fooocus! I missed some of SD1.5 checkpoints.
Before using Fooocus (when I was using Automatic1111) I managed to create a consistent character using a particular SD1.5 checkpoint, after migrating to Fooocus and losing that checkpoint I lost the ability to continue producing more images of that character. It was a petty.
Your video gave me hope to go back to my early days with using Stable Diffusion. I can’t wait to start using SD1.5 checkpoints right now after posting this comment.
Thank you very much!
You may not get similar enough results but I honestly have not tested enough to compare. If you find it works well would love to hear about it.
I have been using this technique for a long time, I make several attempts on the same seed, and choose the one I am satisfied with
Awesome tut. Thanks.
You're welcome!
Now we need a video on how to run VAEs especially SD1.5 Ones.
quando uso essa opçao o face swap nao fuciona
Can I Use this trick in colab version?
Thank you, but what is with Character-Embeddings (textual inversions) wich was trained in SD 1.5 512x512 Resolution?
Posted in pateron but thought this might be better.. Did you notice Fooocus updated and now has "playground 2.5" in the presets? I have no clue what that is and was hoping you had some info on it.
can anybody tell me please can i use this model in comm use? and what abt style?
hey kleebz appreciate it cause lots of new models on 1.5 coming with better images but when i do face swap with upscale 1.5x it shows me this"Image is too large. Directly returned the SR image. Usually directly return SR image at 4K resolution yields better results than SDXL diffusion" and it returns the same img as in upscale 1.5x
Are you only upscaling once? Because I know that once you upscale if once it will not do another upscale using SD and just do a standard fast upscale.
@@KLEEBZTECH ok but first i mask the face and do invert mask so my face is consistent and outside of it changes in all the direction in outpaint and after that i send this image to upscale 1.5x and swap face of the img in image prompt and do img prompt and vary upscale cause i found this way better to get accurate face swap
@@Designing-hc5pz If you are outpainting that will increase the size of the image which Fooocus will only upscale to a certain size so if the image is already large it probably won't do that type of upscale.
@@KLEEBZTECH ok got it thx 👍👍
You can use Fooocus-MRE branch. It supports SD1.5
Supports it natively and not as a refiner? I don't use the MRE branch myself since it has not been updated in some time it seems.
Thank you, thank you! You are the best.
awesome! I never knew that about the 1.5 and LoRA's! Thank you. Does it matter what Jug version I use. I see they have the X version. Think I should also make a file for the XL's too?
@@TomLally you can use any. I am not a big fan of X myself.
Is it possible to generate seamless textures with Fooocus?
Sounds like a challenge lol
No it is not.
Hello i'm a beginner in AI stuff and use Fooocus just for fun, so what's the benefits of using SD 1.5 ?
Well there could be LoRAs that are made for 1.5 that you could want to use and some people like 1.5 but I personally don't use it myself and would stick with SDXL personally. If you were using a UI that fully supported just 1.5 then you would be able to run it on a lower end system.
Hello, is there a feature similar to v2 that automatically generates many beautiful creative backgrounds
I am not sure I fully understand the question. Are you referring to the style Fooocus V2?
@@KLEEBZTECH V2 is a prompt word for the screen, with little change in the background. What I mean is whether there is a function similar to generating various backgrounds
I’m trying to find out how much does this cost but I’m not finding an answer anywhere 😢 please help
How much does what cost? Fooocus is open source and does not cost anything. Be aware that there are scam sites that will try to trick you into thinking you need to pay.
Hey! Ty for idea with folders! Was manually renaming loras....)) Question here) Since I switched to 1.5 refiner seems like image prompts like pyracanny and faceswap features are not working. Am I doing smth wrong or you cant use them with 1.5 refiner?
I think they only work when the SDXL base model is active. So if you switch to 1.5 very early it will not be very useful.
Was ist bitte der Unterschied zwischen SDXL Model und SD 1,5 ?
SDXL is newer and also trained at double the resolution as SD 1.5 was. That is the simple answer.
thank you
You're welcome
how i can do img2img in fooocus please ?
There is the image prompt section at the bottom. I have several videos on using those different features.
Its buggy when using img2img, but not bug when using pnyrna or smthing like faceswap, and inpainting @@KLEEBZTECH
Do you think SD 3 will be available through it ?
I think there is a good chance. Although I do not expect it to happen immediately.
@@KLEEBZTECH I just think this is the easiest way to use SD and how awesome it would be to use it through Fooocus.
100% agree.
Thanks
Welcome
Bro, do you know where I can find a realistic voice generator in all languages and free? I would be grateful if you could help me
Check out Elevenlabs. elevenlabs.io/
@@KLEEBZTECH thank youu bro
@@KLEEBZTECH Is there a way to use it without paying? Is the same thing done in Google Colab image generator?
@@KLEEBZTECH How do I use it in an unlimited way?
As for something unlimited for free I don't know of any. This you can use for free but there is a limit. But it works very well.
Fooocus is heavy on my machine even with 32gb Ram and 8gb video card 😢
Try recording video at the same time 😢. Why I am setting up a different computer to record with now.
we want SORA please for gnrt vd
TL;DR no you cannot use 1.5. A "refiner" is not the same thing as using a base model.
If you put the switch to 0 it's almost the same thing in FOOOCUS 💖💖. However you can use any base model as a refiner. Try yourself!
A setting logic can be:
Forced Overwrite of Sampling Step=25 or whatever
Forced Overwrite of Refiner Switch Step=1
With these settings you bypass all default settings.
Use SPEED mode and change everything in Developer Debug Mode including the Latent size which can be 512x512 or less for speed.
Consider that you lose 1 processing step due to the Switch Step=1 so increase the Sampling Steps accordingly with the switch value and with only these parameters you configure everything to use the SD1.5 models 😉👍
@@AnimAI-LoraI’ve tried this multiple times, the results are always subpar compared to using 1.5 model in A1111. There is a slight resemblance but poor quality. Also it takes much longer to generate the image. So there is really no point.
@@WhySoBroke I would like to start by saying that FOOOCUS should be used as intended by its creator 'lllyasviel' and as expressly specified by him it is finely calibrated to work within certain parameters.
💖 For the rest you are absolutely right, 👍 because everything depends on the first Sampling Step or steps that you generate from the Base Model which leaves the Refiner with a Latent Image which will be used as a starting point..
It also depends on which Model you used as a Base both for the quality or type of result and for the speed of execution.
To obtain the best possible result it is best to use two similar models such as realcartoonXL_v6 and realcartoonAnime_v10 which are from the same family but XL and 1.5.
Furthermore, as Base Model you can use the smallest possible models (4-6GB) to reduce VRam requests and Pruned 1.5 models as refiners (2Gb).
Set Forced Overwrite of Refiner Switch Step to 0-1.
With an optimal swap setting on the SSD it is possible to generate a 512x512 image in 20 seconds using only 2GB of VRAM.
For quality, consider the differences in settings between Automatic and FOOOCUS in the way they generate (ADM, CFG scale, and dozens of other parameters).
In the end it's luck that decides if you've made a good setting!😉
👀 However, consider that we are talking about a fallback solution for those who do not have sufficient hardware resources and do not know how to use Automatic1111 or do not know how to configure the use of VRAM, therefore they are willing to sacrifice something in order to generate with FOOOCUS which is also easier to use.😊
In conclusion, I also use FORGE and COMFYUI with both XL and 1.5 Models, but to use them with limited hardware resources (GTX960 2Gb VrRam) you need to make appropriate settings which are not within everyone's reach.
I hope it has been useful to you and that you don't take it as "the solution", because if you have done a bit of testing you will have realized that the results vary greatly even for a single comma and for each user with different configurations, hardware, software , releases and so on...
💖💖💖 A hug to everyone 😍
@@AnimAI-Lora but you cannot put the switch to 0 :-(
support comment