Just an FYI it's quite possible this fork may break down the road and some functions like creating checkpoints of your face model may not work. Hopefully the developer of FaceSwap Lab makes an effort to make it fully compatible with Forge.
Спасибо тебе, брат! Ты - лучший!!! Теперь мне вообще не надо запускать sd теперь я все делаю во flux. И ты единственный во всем ютубе кто об этом рассказал!
you may have to launch it a couple times for it to download the dependencies properly. I just did a fresh install again as I was moving some files around and it took 2 restarts to get it to show.
Why don’t you have AE, CLIP_I, T5XXL displayed? How does it work without them on this model? and what does AUTOMATIC(FP16 LORA) give instead of just AUTOMATIC?
I'm using the all in one FP8 model in this video so no need to load the text encoders and VAE. As for the FP16 Lora setting, you switch to that when using a lora which I mentioned in 4:06 otherwise Forge will keep loading the lora and it will take longer. This setting keeps it in memory in a sense.
Sure, just set the cfg to 1.1 and the negative prompt area will be enabled. But I can’t say for sure if it really makes a big difference if anything it’s slight changes.
why is this better than ReActor? It works well and has upscale also. NOTE: Codeformers changes YOUR face to another [reduces coherence] compared to GFGAN.
I just checked out reactor from a suggestion in the comments, always heard about it but never bothered. So far I think they are very similar but I'd have to play around with reactor a little more. Better to have options right?
@@MonzonMedia yes sir, i noticed faceswap has MORE options for POST. But its fine, i see it requires installation that you can do easier with reactor, but hey, its cool.
@@MonzonMedia the only downside to THESE apps is that they use the ONNX 128, which people say its very LOW quality.... any words on that? Here’s a breakdown of what it means: ONNX: The model is stored in the ONNX format, making it easy to run on various frameworks or platforms that support ONNX. This helps in ensuring compatibility and allows the use of optimized libraries. 128: This typically refers to the input resolution of the model, meaning the images used in the model are resized to a resolution of 128x128 pixels. Lower resolutions, such as 128, are often used to speed up processing, especially when performing complex tasks like face swapping. In face-swapping models, the input size impacts the detail and quality of the output, with higher resolutions (like 256x256 or higher) offering more detail but requiring more processing power. The ONNX 128 model is likely a version of a face-swapping model optimized for faster performance while maintaining reasonable quality due to the 128x128 input resolution.
Anyone able to run faceswap on img2img ? It always gets me 2 errors : 2024-10-06 11:59:18,655 - FaceSwapLab - ERROR - Failed to post-process 'tuple' object has no attribute 'copy' , and : FaceSwapLab - ERROR - Failed to swap face in postprocess method : 'tuple' object has no attribute 'height' . Anyone had success ?
@@MonzonMedia 4x-UltraSharp,4x-UltraSharp15_4x_foolhardy_Remacri60_lollypop25,4x-UltraSharp25_4x_foolhardy_Remacri75,4x_NMKD-Superscale-SP_178000_G,4x_fatal_Anime_500000_G All of my zoom functions are ineffective... I don't know why~
I don't think the model plays a role. Yes. I've been using face swap lab for a while,it was better than roop.but reactor surpassed it at some point imo.
I have a landscape photo, can I change it as if it was taken with snow or sunset? How should I do it. Especially in Flux quality, it would be very nice to do this.
Face fusion is strictly for face swapping and it can do it for video. I do like it's ease of use. However FaceSwap Lab extension within forge has the advantage of generating an image of your choice and doing the swap in real time. Where as with Face fusion you have to still generate an image then import a source image to use.
@@MonzonMedia Thank you. I know face fusion I could have mentioned that I meant more the quality in my question. It's not a real problem having to import the generated picture. At least for me :) but quality is most important. :)
This is using the latest version of Forge, just did an update this morning and as I mentioned in the video, it's a fork of the original to work with the latest version of Forge.
Hello, I have a question. What is the function of these three elements: ae.safetensors clip_l.safetensors t5 xxl fp16.safetensors in flux forge? I have been able to generate images without them and also with them. If I don't use them, do I lose image quality?
That's a loaded questions but hope this video helps. th-cam.com/video/qt-TeQVpvxg/w-d-xo.htmlsi=wfdEaKuvZNr52BqM Basically those are only needed with models that have the unet as part of the checkpoint. For example if you are using Flux Dev FP8 or NF4 they have those text encoders and the VAE baked in. But the GGUF models require those things to be loaded.
@@MonzonMedia I'm not sure but I think in NF4, these three items were useful if you wanted to use loras. Is this true or is it not necessary to use them to use loras?
You only need them with the GGUF models. If you are using NF4 or FP8 model, you can just load the lora normally. Note that some Lora's won't work with NF4, it's still experimental. Lora's work fine with FP8 or any of the GGUF models. What GPU do you have? I only have 8GB VRAM and I can run the FP8 model with no issues even with Lora.
@@MonzonMedia Thanks for your reply. I have 8 vram. Yes, you can use the FP8 one, in my case it takes longer to generate images, I continue with NF4. It is good to know that I no longer need to load those three elements to use loras.
That's for another video 😊 however with Spaces Photomaker V2, it doesn't keep my likeness as much as Faceswap lab, even if I adjust settings. But I do like the controls it has. FaceSwap Lab is quick and easy to use for most of my needs.
i followed your instructions but does not work on forge, i get this error message in my cmd window when loading SD... (ssl.SSLCertVerificationError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:997)) any thoughts on how to fix?
Why not? ☺️ it’s a good way to use a face of your choice to generate an image. Family, friend or even clients. I do photography on the side and often created fantasy characters with my clients face
these videos are all not suitable for beginners. no info which automatic is used because the one in the video is different from vlad's version. no info where flux comes from where you have to install it etc.. so basic info what is used would be enough.
It made my forge stop working ehehheeh Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)] Version: f2.0.1v1.10.1-previous-527-g720b80da Commit hash: 720b80daea9070aa7896d8ce2c4a6cf8daf927cf
It can be tricky. I cannot get any of these extensions (FSL, ReActor, Roop) to even install in my forge. All of them show that a ModuleMissing Error, and the module is there. There is just something wrong with my Forge install as it seems to be my problem only. I am sure you are having a similar issue. Try Deleting the venv folder then it will add it back when you run again (this is the next best thing to a clean install). Good luck.
@@MonzonMedia Actually I have not figured it out. (I have just broken A1111 numerous times and that seems to be the most consistent fix.) I have tried Forge-AMD using both DirectML and Zluda, neither will install any swapping extension. However with A111 no problem. Same python (different venvs), fresh installs, switch drives... I'm stumped and bummed, I wanna use Flux and FSL.
Just an FYI it's quite possible this fork may break down the road and some functions like creating checkpoints of your face model may not work. Hopefully the developer of FaceSwap Lab makes an effort to make it fully compatible with Forge.
Could you please create a google collab for this?
Долго искал инструкции как запустить у себя Face swap, ни один не сработал кроме этого. Спасибо автору этого канала!
You're welcome and glad it was helpful!
Спасибо тебе, брат! Ты - лучший!!! Теперь мне вообще не надо запускать sd теперь я все делаю во flux. И ты единственный во всем ютубе кто об этом рассказал!
Reactor работает спокойно
I think you are pretty good looking. No ugly there!
😊you are too kind my friend!
@3:20 Man , that comment got me 😂 you got a sub buddy
Glad to entertain! 😬😊 Appreciate the support!
If there are two people in a picture, how to swap the two different faces?
I tried it here and it works fine. Thanks! :)
Great to hear! Thanks for letting me know.
I cant see faceswap lab on main page and appear in installed extension. Please help
you may have to launch it a couple times for it to download the dependencies properly. I just did a fresh install again as I was moving some files around and it took 2 restarts to get it to show.
Great video! Again! But you mentioned faceswap, but i didnt notice that you explained how to image swap from imgtoimg, how do I do that?
Why don’t you have AE, CLIP_I, T5XXL displayed? How does it work without them on this model? and what does AUTOMATIC(FP16 LORA) give instead of just AUTOMATIC?
I'm using the all in one FP8 model in this video so no need to load the text encoders and VAE. As for the FP16 Lora setting, you switch to that when using a lora which I mentioned in 4:06 otherwise Forge will keep loading the lora and it will take longer. This setting keeps it in memory in a sense.
Perfect results!
Awesome!
I can't see the tab "Face Swap Lab". Why?
Shut down and restart, might have to do it a couple times
@@MonzonMedia it's the same
Simple & clear tut. Thank you👌
You're welcome my friend!
why are my faceswaps blurry?
you can use negative prompt with flux???
Sure, just set the cfg to 1.1 and the negative prompt area will be enabled. But I can’t say for sure if it really makes a big difference if anything it’s slight changes.
why is this better than ReActor? It works well and has upscale also. NOTE: Codeformers changes YOUR face to another [reduces coherence] compared to GFGAN.
I just checked out reactor from a suggestion in the comments, always heard about it but never bothered. So far I think they are very similar but I'd have to play around with reactor a little more. Better to have options right?
@@MonzonMedia yes sir, i noticed faceswap has MORE options for POST. But its fine, i see it requires installation that you can do easier with reactor, but hey, its cool.
@@MonzonMedia the only downside to THESE apps is that they use the ONNX 128, which people say its very LOW quality.... any words on that?
Here’s a breakdown of what it means:
ONNX: The model is stored in the ONNX format, making it easy to run on various frameworks or platforms that support ONNX. This helps in ensuring compatibility and allows the use of optimized libraries.
128: This typically refers to the input resolution of the model, meaning the images used in the model are resized to a resolution of 128x128 pixels. Lower resolutions, such as 128, are often used to speed up processing, especially when performing complex tasks like face swapping.
In face-swapping models, the input size impacts the detail and quality of the output, with higher resolutions (like 256x256 or higher) offering more detail but requiring more processing power.
The ONNX 128 model is likely a version of a face-swapping model optimized for faster performance while maintaining reasonable quality due to the 128x128 input resolution.
I can see you have controlnet installed. Is it for Flux?
It’s the default one that comes with forge for SD and SDXL. Forge ControlNet for flux is being worked on. See 6:30
hmm, so video2video isnt possible with that?
Nah just for images but face fusion works with video but very vram hungry th-cam.com/video/WqpcAxoPhL0/w-d-xo.htmlsi=AxJrR7vMMn0q_vLG
What's this like vs Reactor on the flux version?
Actually they are very similar, FaceSwap Lab just has a few more options.
@@MonzonMedia 😅😅 I just started watching your video to install it as you replied!
Can't help but wonder if you happened to use my Bruce Lee Flux Lora for the Bruce Lee images 😊😊
Nah that was a prompt. But I’d love to check out your lora! Love Bruce!
Anyone able to run faceswap on img2img ? It always gets me 2 errors : 2024-10-06 11:59:18,655 - FaceSwapLab - ERROR - Failed to post-process 'tuple' object has no attribute 'copy' , and : FaceSwapLab - ERROR - Failed to swap face in postprocess method : 'tuple' object has no attribute 'height' . Anyone had success ?
Anyone have an idea why sometimes the image gets blurry?
Why is the clarity of the face in the picture I replaced very poor, only the face is poor.
Are you using the upscale option?
@@MonzonMedia 4x-UltraSharp,4x-UltraSharp15_4x_foolhardy_Remacri60_lollypop25,4x-UltraSharp25_4x_foolhardy_Remacri75,4x_NMKD-Superscale-SP_178000_G,4x_fatal_Anime_500000_G
All of my zoom functions are ineffective...
I don't know why~
I have actually been doing this with reactor in Forge works fine
Is it compatible with Flux models?
I don't think the model plays a role. Yes. I've been using face swap lab for a while,it was better than roop.but reactor surpassed it at some point imo.
@@MonzonMedia I have been using with the v4 model with no problem at all
I’ll check it out, never tried it because I liked this one and I also use face fusion. 👍🏼
@@MonzonMedia I can confirm that Reactor works well with Flux.
I have a landscape photo, can I change it as if it was taken with snow or sunset?
How should I do it. Especially in Flux quality, it would be very nice to do this.
You can use inpainting.
Any difference to Reactor?
Pretty much the same thing. This one does have an option to save the faces as checkpoints but it no longer works for me.
So does this work better than FaceFusion?
Face fusion is strictly for face swapping and it can do it for video. I do like it's ease of use. However FaceSwap Lab extension within forge has the advantage of generating an image of your choice and doing the swap in real time. Where as with Face fusion you have to still generate an image then import a source image to use.
@@MonzonMedia Thank you. I know face fusion I could have mentioned that I meant more the quality in my question. It's not a real problem having to import the generated picture. At least for me :) but quality is most important. :)
The ext doesnt work on webui-force Version: f2.0.1v1.10.1-previous-527-g720b80da. It might work in lot older version.
This is using the latest version of Forge, just did an update this morning and as I mentioned in the video, it's a fork of the original to work with the latest version of Forge.
This is giving me no changes as the original image is still the same reactor seems to work but not this
Works fine for me, just used it yesterday. Reactor is pretty much the same so you are not missing out on anything.
this does not work in webui forge
@@genin69 clearly it does, you think I’d do a video on something that doesn’t work?
@@MonzonMedia i just followed ur step by step and it doesnt work. The tab is there but it doesnt open at all. Restarted several times
Reactor works tho
Do you have the latest version or forge and updated it?
@@MonzonMedia last updated last week Wednesday
Hello, I have a question. What is the function of these three elements: ae.safetensors clip_l.safetensors t5 xxl fp16.safetensors in flux forge? I have been able to generate images without them and also with them. If I don't use them, do I lose image quality?
That's a loaded questions but hope this video helps. th-cam.com/video/qt-TeQVpvxg/w-d-xo.htmlsi=wfdEaKuvZNr52BqM Basically those are only needed with models that have the unet as part of the checkpoint. For example if you are using Flux Dev FP8 or NF4 they have those text encoders and the VAE baked in. But the GGUF models require those things to be loaded.
@@MonzonMedia I'm not sure but I think in NF4, these three items were useful if you wanted to use loras. Is this true or is it not necessary to use them to use loras?
You only need them with the GGUF models. If you are using NF4 or FP8 model, you can just load the lora normally. Note that some Lora's won't work with NF4, it's still experimental. Lora's work fine with FP8 or any of the GGUF models. What GPU do you have? I only have 8GB VRAM and I can run the FP8 model with no issues even with Lora.
@@MonzonMedia Thanks for your reply. I have 8 vram. Yes, you can use the FP8 one, in my case it takes longer to generate images, I continue with NF4. It is good to know that I no longer need to load those three elements to use loras.
Is this available for NSFW content?
no, only SFW
no, there is also another way: in Forge Spaces we have Face Swap (down below in the list of spaces)
That's for another video 😊 however with Spaces Photomaker V2, it doesn't keep my likeness as much as Faceswap lab, even if I adjust settings. But I do like the controls it has. FaceSwap Lab is quick and easy to use for most of my needs.
@@MonzonMedia :)
Doesn't look much better than any of the other face swappers. All they do is swap it with a face that kinda looks like the original, but not really...
i followed your instructions but does not work on forge, i get this error message in my cmd window when loading SD... (ssl.SSLCertVerificationError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:997)) any thoughts on how to fix?
why do you want to swap a face?
Why not? ☺️ it’s a good way to use a face of your choice to generate an image. Family, friend or even clients. I do photography on the side and often created fantasy characters with my clients face
Topaz Gigapixel perfect upscaler
Yeah I have it as well but the upscaler in the video is to upscale the face quality since the face swap model is low resolution.
@@MonzonMedia aaaa ok )
No, it's outdated.
these videos are all not suitable for beginners. no info which automatic is used because the one in the video is different from vlad's version. no info where flux comes from where you have to install it etc.. so basic info what is used would be enough.
It's not intended for beginners, I have other videos for that.
It made my forge stop working ehehheeh
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f2.0.1v1.10.1-previous-527-g720b80da
Commit hash: 720b80daea9070aa7896d8ce2c4a6cf8daf927cf
It can be tricky. I cannot get any of these extensions (FSL, ReActor, Roop) to even install in my forge. All of them show that a ModuleMissing Error, and the module is there. There is just something wrong with my Forge install as it seems to be my problem only. I am sure you are having a similar issue. Try Deleting the venv folder then it will add it back when you run again (this is the next best thing to a clean install). Good luck.
@@AdamNewcombMumbleTrash yeah , reinstalled all from scrap and worked heheh
I was going to say to try a fresh install but I see you figured it out. 👍🏼
@@MonzonMedia Actually I have not figured it out. (I have just broken A1111 numerous times and that seems to be the most consistent fix.) I have tried Forge-AMD using both DirectML and Zluda, neither will install any swapping extension. However with A111 no problem. Same python (different venvs), fresh installs, switch drives... I'm stumped and bummed, I wanna use Flux and FSL.