Good video, but to make it a true test you should try the same face image on 4 different images: 1. A front view photo (like you did here) 2. A side/3 quarter view photo. 3. A photo with a different facial expression. 4. A photo where the lighting doesn't match at all. The image you tried to swap here already matched really well as is in terms of angle and lighting. You could have just cut her face out and pasted it on top for a similar result you know...
Thank you for your efforts. My question is that when I try to swap a real photo and not a photo produced by ai, the results are not very good. Do you have any suggestions for this (resolution of the photo, crop rate, shooting angle, etc). I also wonder if there is a more successful biological method you know for the faceswap event.
Hello. This is one of the points that I made in the video. In my honest opinion....Fooocus is not that great with real photos for face swapping. I have had vastly different results from just OK to very poor indeed. Only on a couple of occassions have I been pleasantly surprised with the result. A face photo, similar to the one that I used in the video does tend to work the best though. Inpainting is probably the best method. Hope that helps.
@@FoxtonAI Thank you, so is there a method or an external program that you use that is more successful in this subject (about real life photos). I see some posts on instagram, they are really very successful, they can reflect the person without any loss.
U got any best same face replacement option ?? M also search this past months m getting very bad results need to match same jawline n same face with expression I need to add Already generated image swap@@cemilhaci2
I too have the same issue, probably because I’ve got a head like a smashed avo 😂 anyhoo, if I do a face swap with the missus head it is spot on, uncanny in fact. I tried the faceswapper site with my head and had absolutely no luck there either. Another TH-camr made the statement that not all faces are suitable for swapping I’m starting to believe him. Good luck with it, if you succeed lemme know ✌🏻
HI. I like the way you teach, you;'e quite precise and you give it thought. Thank you. Curious, I have Forge UI for Flux and have downloaded many models for it. New to Fooocus and I intend to install it locally. I see that it also needs models. Would Fooocus need to have a separate folder for its models or is there a way to share them between the two. I'm trying to watch my Hard Drive Space. Cheers.
Thanks for your kind words. Glad the video was helpful. Yes you can share the models. The easiest way is to download the checkpoint models to your Forge UI folder (Forge\webui\models\Stable-diffusion). Then in Fooocus, open in Notepad, the config.txt file (Fooocus\Fooocus\config.txt). At the top of that file, just change the "path_checkpoints" location to your Forge UI folder. Hope that helps. These Fooocus videos may be of help to you: Fooocus: Install in 2 Steps - Latest Version + Dark Theme: th-cam.com/video/0oOUwVrKAC0/w-d-xo.html Add Custom Image Aspect Ratios and Resolutions in Fooocus: th-cam.com/video/4HzUrBx0c44/w-d-xo.html
Face swapping a 'photo' to another 'photo' may be what most people are interested in - but I find it very limiting, considering the possibility of creating individual art styles. Unfortunately there are not that much good tutorials about how one can put a real person's face from one or more source photos into a given target image of, say, cartoon or comic or oil painting or anime style - without explicitly training a LoRA for it. E.g. for creating a personalized children's comic book where the child is the hero. Is any of the methods you tested usable for this use-case?
Hello. I agree with you. Swapping photo to photo is a lot more difficult. I remember that ReActor gave me pretty decent results with realistic images. You mention an additional difficulty, on top of a photo to photo face swap, which is to swap the face onto the target image in the same style as the target image. Most face swap extensions/tools will have difficulty doing this, without additional tweaking. However, again if my memory serves me correctly, InstantID was capable of doing this, particularly when the target image was in a cartoon style. Maybe I'll take a look at all the current face swap extensions/tools, to do a comparison of which ones are best for which types of face swap. By the way...creating a personalised children's comic is a great idea. Hope this helps.
The original image was generated previously. If you go to your output images folder, there will be a log.html file in there. Open that, and you'll see all of the images that you generated for that day, together with the metadata and a copy button. Copy and paste that into the text prompt box, and the button will change smartly to "Load Parameters", press it to populate the parameters. By the way...you can also save metadata directly to the image, by selecting that option in your settings. Hope that helps.
With Fooocus, the only way that I know of, is to do it in two stages. Swap one face, and then use that image to swap the second face onto. If you are also using Automatic1111, you may want to take a look at ReActor, which does multi-face swapping. Hope that helps.
Good video, but to make it a true test you should try the same face image on 4 different images:
1. A front view photo (like you did here)
2. A side/3 quarter view photo.
3. A photo with a different facial expression.
4. A photo where the lighting doesn't match at all.
The image you tried to swap here already matched really well as is in terms of angle and lighting. You could have just cut her face out and pasted it on top for a similar result you know...
Thanks for your suggestions.
Thanks! A good faceswap is so difficult...
Agree!
Great tutorial, THANK YOU, that's what I was looking for.
Thank you very much for your comment. Really glad you found the video helpful.
Thank you for your efforts. My question is that when I try to swap a real photo and not a photo produced by ai, the results are not very good. Do you have any suggestions for this (resolution of the photo, crop rate, shooting angle, etc). I also wonder if there is a more successful biological method you know for the faceswap event.
Hello. This is one of the points that I made in the video. In my honest opinion....Fooocus is not that great with real photos for face swapping. I have had vastly different results from just OK to very poor indeed. Only on a couple of occassions have I been pleasantly surprised with the result. A face photo, similar to the one that I used in the video does tend to work the best though. Inpainting is probably the best method. Hope that helps.
@@FoxtonAI Thank you, so is there a method or an external program that you use that is more successful in this subject (about real life photos). I see some posts on instagram, they are really very successful, they can reflect the person without any loss.
u can use my workflow, but u have to use comfyui for that, results are great for human faces,
U got any best same face replacement option ?? M also search this past months m getting very bad results need to match same jawline n same face with expression I need to add Already generated image swap@@cemilhaci2
I too have the same issue, probably because I’ve got a head like a smashed avo 😂 anyhoo, if I do a face swap with the missus head it is spot on, uncanny in fact. I tried the faceswapper site with my head and had absolutely no luck there either. Another TH-camr made the statement that not all faces are suitable for swapping I’m starting to believe him.
Good luck with it, if you succeed lemme know ✌🏻
Great video and tips!
Thank you very much! Glad you found the video helpful.
HI. I like the way you teach, you;'e quite precise and you give it thought. Thank you.
Curious, I have Forge UI for Flux and have downloaded many models for it. New to Fooocus and I intend to install it locally. I see that it also needs models.
Would Fooocus need to have a separate folder for its models or is there a way to share them between the two. I'm trying to watch my Hard Drive Space.
Cheers.
Thanks for your kind words. Glad the video was helpful.
Yes you can share the models. The easiest way is to download the checkpoint models to your Forge UI folder (Forge\webui\models\Stable-diffusion). Then in Fooocus, open in Notepad, the config.txt file (Fooocus\Fooocus\config.txt). At the top of that file, just change the "path_checkpoints" location to your Forge UI folder. Hope that helps.
These Fooocus videos may be of help to you:
Fooocus: Install in 2 Steps - Latest Version + Dark Theme:
th-cam.com/video/0oOUwVrKAC0/w-d-xo.html
Add Custom Image Aspect Ratios and Resolutions in Fooocus:
th-cam.com/video/4HzUrBx0c44/w-d-xo.html
@@FoxtonAI You are awesome! Really good tutes. They are very relevant and eclectic. Thank you! Subscribed for certain.
Just use stability matrix..
how come none of the methods beyond the base face photo, gave her green eyes?
It's a strange world how the face is swapped in from an image base, it's easy to fix, by just adding "green eyes" in an additional text prompt.
Face swapping a 'photo' to another 'photo' may be what most people are interested in - but I find it very limiting, considering the possibility of creating individual art styles. Unfortunately there are not that much good tutorials about how one can put a real person's face from one or more source photos into a given target image of, say, cartoon or comic or oil painting or anime style - without explicitly training a LoRA for it. E.g. for creating a personalized children's comic book where the child is the hero.
Is any of the methods you tested usable for this use-case?
Hello. I agree with you. Swapping photo to photo is a lot more difficult. I remember that ReActor gave me pretty decent results with realistic images. You mention an additional difficulty, on top of a photo to photo face swap, which is to swap the face onto the target image in the same style as the target image. Most face swap extensions/tools will have difficulty doing this, without additional tweaking. However, again if my memory serves me correctly, InstantID was capable of doing this, particularly when the target image was in a cartoon style. Maybe I'll take a look at all the current face swap extensions/tools, to do a comparison of which ones are best for which types of face swap. By the way...creating a personalised children's comic is a great idea. Hope this helps.
How to reduce the head size of the swap head?
It will autosize.
what model are used in this?
Hello. In the video, I used the JuggernaultXL checkpoint, but you can use others, depending on your preferences and needs.
Wait…. Where In the first example does the meta data get copied from, also - will the button auto change to load parameters? Never seen this before
The original image was generated previously. If you go to your output images folder, there will be a log.html file in there. Open that, and you'll see all of the images that you generated for that day, together with the metadata and a copy button. Copy and paste that into the text prompt box, and the button will change smartly to "Load Parameters", press it to populate the parameters. By the way...you can also save metadata directly to the image, by selecting that option in your settings. Hope that helps.
@@FoxtonAI ahh thanks, makes sense , I was running on colab so need to save that prior, by opening it before i quit session.
@@FoxtonAI This is something I had no idea... Looking at my folders now with all that metadata. Perhaps the best piece of knowledge thus far
How to swipe multi face on one pic?
With Fooocus, the only way that I know of, is to do it in two stages. Swap one face, and then use that image to swap the second face onto.
If you are also using Automatic1111, you may want to take a look at ReActor, which does multi-face swapping.
Hope that helps.
can flux be in focus?
maybe, but probably not.
Unfortunately not...current options are ComfyUI, online and SD Forge (NF4 model).
Thanks a lot !
Thanks for your comment. Glad you liked the video.