Stable Diffusion - Inpainting with Fooocus - Don't Regenerate, Fix!

Поділитися
Вставка
  • Опубліковано 15 гру 2024

КОМЕНТАРІ • 93

  • @xainudin108
    @xainudin108 10 місяців тому +1

    Wow, this is amazing! I learned a lot from this tutorial on how to use inpainting to fix or modify images generated by Stable Diffusion. The examples you showed were very impressive and inspiring. I especially liked how you improved the details of the face and hands. This is a very useful and creative tool for digital artists and creators. Thank you for sharing this video!

  • @lexaunculpt
    @lexaunculpt 10 місяців тому +3

    This is great. Could use inpaint with a lora to replace a face?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +2

      You should be able to. I just double checked with one of my custom ones and was able to use the regular inpaint and then change the prompt with the trigger word for the lora and it worked.

  • @jeffreypodcasttv
    @jeffreypodcasttv 10 місяців тому +4

    make a video showing how to fix the hands, they always come out with defects, I love your videos, I really learned a lot with you. thanks.

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      Actually something I have on my list. 👍

    • @shacharperl5217
      @shacharperl5217 10 місяців тому

      😢ס😮בב😮😮את את 😅😮😅😮סדק😊😅😊😅😅😊😅😅😅😅😊😅😅​

  • @serviodiaz4425
    @serviodiaz4425 9 місяців тому +1

    Thanks, great content.

  • @mariondroguet7944
    @mariondroguet7944 10 місяців тому +1

    Fantastic video!
    Already tested this and it IS amazing! One question - is there a way to use inpaint or other option there to insert a *specific item* to a photo?
    I mean to upload photo/s of a specific shirt, necklace, skirt and inpaint them to a target image?
    Great video, thanks a lot! ❤

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +2

      Not really that I can think of. You could possibly get something similar by using the image prompts combined with inpainting. You do need to enable that option in the Developer Debug Mode. If you were using a LoRA with trained content that would probably be another way.

  • @fabiotgarcia2
    @fabiotgarcia2 10 місяців тому +1

    I´d love ro see a tutorial about how to do image for fashion advertising with beautiful places, beautiful models.

  • @fabiotgarcia2
    @fabiotgarcia2 10 місяців тому +2

    Excellent. Is it possible to put a real dress or shoes in a model. Like take a photo of a real shoes and apply it to the feet’s model?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      Not exactly. You could use the image prompt feature to influence that sort of thing. But to get the exact same, the only way I could think of is a clothing lora (which I have not messed with yet) or I suppose you could take the photo of a person in the real dress and inpaint the rest of the image.

  • @BM-mg6ku
    @BM-mg6ku 5 місяців тому

    Thank you, what is the best setup to get realistic image?

    • @KLEEBZTECH
      @KLEEBZTECH  5 місяців тому

      That is such a difficult question to answer since everyone's view of realistic is different. But start by looking at some different realistic checkpoints on places like CivitAI. I don't use a specific one since they all have their benefits. Also play with some of the photo styles and try adding things like RAW photo to the prompt which can help for some models.

  • @valentinacevenini8751
    @valentinacevenini8751 10 місяців тому +1

    Hi, thanks for your very useful videos, how did you manage to remove the girl's two braids and put just one? I can not make it. can you help me by giving me some suggestions? Thanks again !

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      If you are talking about the two images near the start, those were just a couple images I generated to start with. I did not change them from two to one. But you could with inpainting if you were to change the prompt to specify one braid and mask both of them. Some things take multiple attempts. I made a quick video to show how and it took a few attempts and had to skip the first few generations but worked: ua-cam.com/video/zJ2LOGDqZeQ/v-deo.html

  • @narbig7
    @narbig7 6 місяців тому +1

    Muchas gracias!!

    • @KLEEBZTECH
      @KLEEBZTECH  6 місяців тому

      If you haven't already I do suggest my other videos on inpainting. I have several that all touch on different aspects I have learned to really improve your inpainting. Like this one. ua-cam.com/video/rMQQx6LLGSo/v-deo.html

  • @lilillllii246
    @lilillllii246 8 місяців тому

    Thanks. Can't I use image prompt(cpds,faceswap,pyracanny) and inpaint at the same time on foocus?

  • @harishgovekar3597
    @harishgovekar3597 10 місяців тому +1

    Awesome video!!!! It will be great if you make a video creating AI Influencers

  • @ibrahim__alsayed
    @ibrahim__alsayed 5 місяців тому +1

    thanks

  • @Alphaai1997
    @Alphaai1997 10 місяців тому +1

    hello sir, i downloaded fooocus with the latest version but the ouput is always error

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      What are you getting for an error in the CMD window?

    • @oscargortez
      @oscargortez 10 місяців тому

      I had that too, then I found out my computer apparently doesn't have a beefy enough graphics card.

  • @jamespogg
    @jamespogg 10 місяців тому

    can you do a video about adding new people to a real image? would love that!

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Adding AI people and things to real photos? I can see doing a video on that probably. I have not added people but I have added other things like other animals to photos I have of my goat. 👍

  • @Muslimman570
    @Muslimman570 3 місяці тому

    what are your specs u keep mentioning its not great and thank you i have learned so much from you

    • @KLEEBZTECH
      @KLEEBZTECH  3 місяці тому

      At the time was a 3070 with 8gb vram.

  • @jkashj1994
    @jkashj1994 10 місяців тому

    Does Fooocus or any other AI allow for uploading of a jpeg of an image and then you can utilize some type of inpainting function to edit/augment the reality of the uploaded image. I know that you can use "reference" image in some AIs, but I would want to alter an existing "analog" image. I'm having a hard time finding this as an option and it surprises me that it's not available somewhere. It would be highly marketable.

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      You can inpaint real photos and it does a pretty good job of blending things. I will say some images it does much better than others.

  • @auzland152
    @auzland152 2 місяці тому

    How does this method differ from the new Enhance check box and Enhance tab they added?

    • @KLEEBZTECH
      @KLEEBZTECH  2 місяці тому

      Inpainting is still its own thing. The enhance feature makes use of inpainting to automatically enhance the images. Inpainting the regular way is going to usually be the better option but using the enhance can reduce the need sometimes in my opinion. But the way I do images I rarely use enhance.

    • @auzland152
      @auzland152 2 місяці тому

      @@KLEEBZTECH If I have an image that is missing feet at the end of the legs, would inpainting help with that? I tried masking the area where the feet should be, but it's adding weird feet with completely different lighting...

    • @KLEEBZTECH
      @KLEEBZTECH  2 місяці тому

      Yes that would be the way. You want to make sure using the default inpainting and not one of the other options usually. And keep your prompt for the full image since it is going to generate based on that. It sounds like you may be using one of the other inpainting options.

    • @auzland152
      @auzland152 2 місяці тому

      @@KLEEBZTECH I was using the modify option so that I could input an additional prompt like "add feet" but wasn't sure if that was the best way.

    • @KLEEBZTECH
      @KLEEBZTECH  2 місяці тому

      Not usually. And if you do use the modify you do not want the full prompt left since it uses that as well for that small area. Just use the default and usually that should work since it usually should put feet on people. If you want to help you could add something about the feet in the regular prompt.

  • @waynescholar4998
    @waynescholar4998 10 місяців тому

    Can you inpaint and use a specific image (like face swap) to replace the area with that specific image? For example if i have a model wearing a jacket and an image of a new jacket i want to put on that model keeping all the details?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      No. You might be able to get close using image prompt set to high stop and weight but not going to get the same clothing.

  • @mysticgatekeeper
    @mysticgatekeeper 8 місяців тому

    As a very new AI user (less than 2 weeks) I've been exploring Fooocus and noticed something at my end regarding character creation. I've been unable to render a person with a full body. At best I see the person from the waist up whether sitting or standing (and yes, they are clothed). Is this how Fooocus works or am I doing something wrong at my end? Thank you.

    • @KLEEBZTECH
      @KLEEBZTECH  8 місяців тому

      That is one of the things you will learn as you get better. Give this video a go. Hopefully it will get you down the right path ua-cam.com/video/YODdFMlZz2Q/v-deo.html

    • @mysticgatekeeper
      @mysticgatekeeper 8 місяців тому

      @@KLEEBZTECH Thanks for replying and for the link. I"ll be reviewing it today.

  • @prohabitat
    @prohabitat 9 місяців тому

    Ser... i have an issue.. i get connection error as soon as i upload an image to inpain sections or if i use a photo for face swap. when i press generate it crashes?? plez welp

    • @KLEEBZTECH
      @KLEEBZTECH  9 місяців тому

      Are you running locally on your own computer or using it online with a service? Any error messages would be useful in determining the issue. If running locally you should get some sort of error in the CMD window for example.

  • @alpenjager3986
    @alpenjager3986 10 місяців тому +1

    Can you please make a tutorial for fixing mutated hands with Fooocus?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      I do plan on making something about AI hands but there is no great fix. But there are some tips I will give when I make it to help reduce the number of inpainting attempts when trying to fix them. I am also open to any tips others have found.

    • @alpenjager3986
      @alpenjager3986 10 місяців тому +1

      @@KLEEBZTECH I think raising the topic of Fooocus Stable Diffusion will have a positive impact on the development of your channel, considering the lack of content creators who provide this information.

  • @dpgedward5947
    @dpgedward5947 10 місяців тому

    I use foocus but I keep getting bad results. Like putting handles on kitchen door, they are all over the place, also always getting extra vertical rails. can you train the program to fix this?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Much comes down to good prompts and some luck. You can also try different models. You can not really train Fooocus itself when it comes to those things. I do suggest when testing out different prompts also using the same seeds to see the impact the changes have.

  • @LEGIT134
    @LEGIT134 10 місяців тому

    As long as you aren’t making Disney characters or something obviously copyright you can sell these images right? Like for a book for example

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Yes usually. But I am not a lawyer. 😀 There is still a lot of legal things to be worked out with them.

  • @tradehut2782
    @tradehut2782 10 місяців тому

    Is there a best checkpoint for inpainting?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Should work fine with any.

    • @akuma2124
      @akuma2124 9 місяців тому

      @@KLEEBZTECH currently only supports models that are SDXL. If you try to run one that isnt supported, the cmd window will say its not supported.

    • @KLEEBZTECH
      @KLEEBZTECH  9 місяців тому

      Correct. You can use others as a refiner. But Fooocus is designed for SDXL. ​@@akuma2124

  • @vrrevolution9183
    @vrrevolution9183 10 місяців тому

    i have troed over and over t get rid or earrings, just makes new ones each time, even with negative prompts... you have any idea how to do that ? i see yours changed earrings too

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      I am going to try it and see. But having a hard time getting any women with earrings... lol.

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      Have you tried using heavier weights in the negative prompt? The negative prompt can be hit or miss with what it will impact but using higher weights might help. Maybe like this: (earrings:1.5), (jewelry:1.5)

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      I did finally get some by requesting earrings. Then I removed that from the prompt and just masked the lower part of the ear and the earring. Mine removed it every time... Just be sure you are using a random seed in case you may have unchecked that box.

    • @vrrevolution9183
      @vrrevolution9183 10 місяців тому +1

      nice if you get a chance shoot a video, i do so much better with visuals! @@KLEEBZTECH

    • @vrrevolution9183
      @vrrevolution9183 10 місяців тому +1

      your tuts are awesome thanks!

  • @keepitshort4208
    @keepitshort4208 10 місяців тому

    The problems I'm having is the white halo around the subject background

  • @cmdr_stretchedguy
    @cmdr_stretchedguy 4 місяці тому

    There is also LORAs for better eyes, faces, expressions, and hands to get better results off the start.

    • @KLEEBZTECH
      @KLEEBZTECH  4 місяці тому

      Yes but I have not found those to work very well. Especially the hands.

  • @bryanbrewer4272
    @bryanbrewer4272 9 місяців тому +1

    very moist

  • @kenmasters734
    @kenmasters734 10 місяців тому

    Somthing is borked with the new focus model update, faces come out with the same stupid half smile and dont want todo different expressions.

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Checkpoint models were recently updated but you can use the previous ones.

  • @Bamigaru
    @Bamigaru 10 місяців тому

    HI! How are you? This video is also perfect as always. I'm trying to create and edit the photos myself as I follow you in your excellent step by step explanations. This way I remember better what does what. I tried to use a photo where the mouth is ajar. I used Inpaint>Improve Detail masking the mouth area and around it to give the Ai more room to maneuver as you say in the video. As a prompt in the Improve Detail box I put "detailed partly open mouth" (I did some tests, putting it as a general prompt + Improve Detail prompt doesn't change anything, the result is always the same), but the result is very fake: the texture of the mouth is different from that of the rest of the face (I also used 2x upscaling before using the impaint, as you suggested and in fact the results are better) and the shape of the mouth does not follow the features of the face. Do you have any advice on how to fix it?

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      For something like that I find doing the whole face works better.

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +1

      Can also check off the developer debug mode and in the inpaint section adjust the denoise strength to fine tune things.

    • @Bamigaru
      @Bamigaru 10 місяців тому

      @@KLEEBZTECH Will it not change the subject's face? because im trying to recreate the same face but with different expressions (with expressions i mean only mouth and eyes sometime)

    • @Bamigaru
      @Bamigaru 10 місяців тому

      @@KLEEBZTECH i'll try with this, thanks!!!

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому

      Another possible idea is to use FaceSwap with inpainting to redo the face...

  • @wernerblahota6055
    @wernerblahota6055 10 місяців тому +1

    👍👍👍👍

  • @meikmeiker2615
    @meikmeiker2615 9 місяців тому +1

    A video that goes into more depth would be great

    • @KLEEBZTECH
      @KLEEBZTECH  9 місяців тому

      I do plan on combining all the tips and techniques for inpainting from all my videos into a more comprehensive one in the near future.

  • @cyril1111
    @cyril1111 10 місяців тому

    But you barely scratched the surface of inpainting here. Why you didn't talk (or maybe you are not aware of) about the advanced Tab > developer debug > Inpaint Tab - Here you can choose your denoise strength, inpaint latent or not, and so many more options that gives you full power on the inpainting!

    • @KLEEBZTECH
      @KLEEBZTECH  10 місяців тому +2

      As I mentioned this is just the basics and will be doing a video on more of the advance stuff soon. I could spend much longer covering everything but then it would be a very long video that most people would not watch or be interested in.

  • @pedro3000
    @pedro3000 9 місяців тому

    Count how many times you say "actually" in this video.

    • @KLEEBZTECH
      @KLEEBZTECH  9 місяців тому +1

      Actually it is about once per minute. 😉

    • @pedro3000
      @pedro3000 9 місяців тому +1

      @@KLEEBZTECH I actually did not count, but I’m sure it’s actually quite a few. Your content is great, you could actually use fewer adverbs and it would actually be more concise 👍

    • @KLEEBZTECH
      @KLEEBZTECH  9 місяців тому

      😂 Working on it. I did count though since it only took a few seconds to check the subtitles and it was about once a minute. 👍

  • @fulldivemedia
    @fulldivemedia 5 місяців тому +1

    thanks