Product Placement Tips For Fooocus Image Prompt/Inpaint (Stable Diffusion)
HTML-код
- Опубликовано: 27 май 2024
- Trying to add a real object or clothing item into a stable diffusion scene? Check out these tips for using fooocus ai’s best feature, its inpaint. Adding a product or any real item that has specific details and logos is impossible without a little extra manipulation when it comes to stable diffusion. While it may seem complicated at first it is actually quite easy.
Photopea Image editor: www.photopea.com/
Photoroom BG Remover: www.photoroom.com/tools/backg...
Adobe Express BG Remover: www.adobe.com/express/feature...
Masb1t Fooocus Fork: github.com/mashb1t/Fooocus
If you want to help support this channel! www.buymeacoffee.com/jumpintoai
Check Out my Full Fooocus Tutorial Playlist!: • Fooocus!
Want the easiest SDXL Stable Diffusion software? Go to the Fooocus download page: github.com/lllyasviel/Fooocus
----
00:00 Intro
00:20 Image Prompt
02:50 Inpaint a Model
07:21 Change Existing Model
10:23 Shoes Example
11:05 Non-Clothing Item
11:51 Bottles-Transparent Items
12:39 Fooocus Fork - Auto Generate Mask
---- 
Наука
Fooocus download page: github.com/lllyasviel/Fooocus
All other links are in the Description
Excelentes tutoriales de Fooocus, tu canal es increible, espero ver más trabajos así, saludos desde Chile 👋
Thanks!
Simply amazing, one of if not the best tutorial in god forsaken YT for this type of work.
Found this at the perfect time ❤
Happy to help!
Great lesson! Thank you so much!
You are very welcome!
thats f* excellent tutorial bro thanks, bless you
Wow, thanks you professor!
thank you for this workshop 😁
Great Video, Loved it
Ty for the awesome videos 🙂
Very useful video.thank you...
Welcome 😊
Great video
so good taste
Is tere a way I can contact you for a private 1 on 1 lesson? I am trying to get ai models for our socks at different positions, locations etc but I couldn't make it work following your tutorial. I am willing to pay for the private tutorial. thank you
Thanks for this film! Would be also nice to learn how to inpaint exact text on an element. Recently I tried a birthday cake with a certain text on it in a style of "creamy cake text".
the best videoi try to find
This is really impressive. I've never tried fooocus yet, but it seems to do a lot of cool things pretty easy. If you don't care about videos yet, would Fooocus be better for editing images?
Thanks for the great video. I only have one problem. Even though my image and the black mask have the same aspect ratio the mask seem to not get placed exactely over the image I want to inpaint. Therefore I just get bad results. In your video you can see that the image and the mask are exactely at the same spot. When I upload both (image and mask) I can already tell its not at the same place. As I said the aspect ratio and the position of the object/mask are exactely the same. What could be the reason for that?
Hi, Very ggod tutorial. thx for this. One question. Did try this method wit jewelry? Rings or necklace?
TLDR - Yes but the results weren't worth the effort.
Due to those being so small and intricate, It's incredibly difficult to get a decent image without a very specific setup. Even then, it doesn't always work.
Most items like that are only practical using the method of repainting the human model wearing it rather than trying to add a model to a piece of lone jewelry. Even then, the AI likes to alter the edges of jewels making them not quite "exact".
@@JumpIntoAI Well, I just had such an experience. I'm waiting for something to change in this matter because I design jewelry in 3D and it would be very useful for me to use AI models for this.
Can you make an installation instruction video? Thank you so much! ! !
I have one that might be a little outdated now (and not very good since it was one of my first videos) but you can check that out here: ruclips.net/video/828wYIp2HAM/видео.html
You can go to the full details of the dress. Did the position change?
I can´t get the "non-clothing" part to work. Only the Inpaint image shows. Alpha also do nothing. Prompt does nothing.
Great video Jump! would you know how you can do this within MidJourney ??
I haven't used midjourney for a bit but I don't believe it has the capabilities of inpainting/outpainting an original image, especially with this precision.
@@JumpIntoAI i've been trying but it seems like you can't, is there any way we can use Fooocus on Macbook -i've been able to get fooocus colab google server but it seem to not work, any tips to make it work ?
great, now you need to make a character turnarround with face and clothes you just invented (rather than text prompt)
If my product is handbag, can use this method?
Can you create this placement for jewelry
Hey brother in part "inpaint a model" when i upload tshirt image at one side and mask image another side and after giving prompt it's not giving results with the exact propmt thsirt is random place on plane background like. But not weared ant human...please help
you did an Awesome tutorial! one thing I'm in doubt:
how could I combine 2 images, imagine I have already created a man but I want to make this man holding a perfume that exists, I got the image from an online store. How could I combine this two images? Is there a way to inpaint and pass a image as reference and do the same like you did with dress to keep the details?
Hello, let's say I have a pose sitting on a chair, when I select my body except the head and try to make a muscular man, yes it makes me muscular, but it poses again, I want it to maintain the current pose. This is possible in improve detail, but I want to make it fully variable, I mean, in the improve detail part, I want to make a man with clothes on without clothes and muscular because there is no intervention in the pose there, is it possible to do this there?
so if i want to use this Auto mask generation feature, i need to delete the lllyasviel version of fooocus and again install mashbit's version of fooocus ?
No, treat it like a separate installation of fooocus.
Install it in its own directory, and then change the config.txt so that it will use the model folders from original fooocus (or wherever you keep your models) so it wont have to redownload them again.
You should try Defooocus, which include Photopea and backround remover in one.
I have actually, a month or more back. But it I had some issues with it and dropped it.
did you use styles or any loras in img prompt and also in inpainting??
I left everything default for this, so the 3 default styles were checked (fooocus v2, fooocus enhanced, fooocus sharp), juggernaut model used, and no LoRAs
@@JumpIntoAI ok thx 👍 and can you tell how can i make juggernaut x as default model in google colab
@@Designing-hc5pz I don't use the colab version but im pretty sure Juggernaut is the default model automatically.
@@JumpIntoAI yup but they released version 9 & version x tomorrow but they didn't update it on collab or any other platform like sagemakker
@@Designing-hc5pz Ok I Understand now, I am not a colab expert but I "think" you can add the new model with this line:
!wget -O 'models/checkpoints/juggernautXL_juggernautX.safetensors' "civitai.com/api/download/models/456194?type=Model&format=SafeTensor&token="
Before starting the notebook, go to the top where it says +Code and then add that line and run it. then start the notebook when its done.
Is it possible to create images of a room from different angles with the same products in interior design using Foocus? Can you make a video about this?
Creating different angles of the same room from scratch isn't going to be possible with stable diffusion currently.
@@JumpIntoAI Do you have any ideas if it is possible in another ai program?
You are supper !
Thank you! 😃
Sorry this may be out of the context of your video. But can we upload the lora file on fooocus colab quickly? Or maybe without having to download the lora first?
A fellow on reddit named eddyizm created a tutorial just for that.
Check that out here: eddyizm.com/blog/44-fooocus-loading-loras-into-google-colab/
Hi, in the advanced settings panel the presets are unchecked, how can I do this?
Each preset is just a collection of settings. Clicking on the realistic will change the model/lora/neg prompt/styles/ resolution/etc. to something specific.
If you have a certain model/lora/resolution/ other settings that you always use, you can create your own preset for that.
It will also download any missing models from those presets, so clicking them for the first time will cause that to happen.
@@JumpIntoAI Thank you
but all my clothes brand are with logos. always.. and fooocus removing my logos. how i can keep the logo as it is?
Unfortunately, you wont be able to generate your clothes logos or any specific minor details. The only way is to start with a real picture of the clothes that's in the shape of someone wearing it (like my example with the dress and the brown t-shirt, having it either on a mannequin or another person). And generate the person around the clothes.
Can i do it on my own images like my pictures
Yep! I used my own photo taken with a phone camera as an example for the t-shirt.
What is the minimum system requirements for fooocuse?
github.com/lllyasviel/Fooocus?tab=readme-ov-file#minimal-requirement
hey buddy just had a question, how to combine two different models in foocus , like if i had an ai model and i want to collab with another model how do i do it please reply >
I would have a LoRA of at least one of those models and a picture of the other and try to inpaint the model with the LoRA into the image with the other. Ideally the image would have 2 people already and just inpaint over the other person. With no LoRA then its dipping into photoshop as faceswap is good but not exact.
@@JumpIntoAI Why LoRA? Which one?
@@hmm-m If he wants 2 very specific People in a scene the simplest way to me would be having an image of one of them (even better if they are with another person in the image) and then using a LoRA created for the second person, and inpainting them into the image (over that second person in the image if there is one)
I've done something similar with character LoRAs.
Assuming this all done in fooocus.
You could just use faceswap on a similar looking model but it can be very hit and miss with some faces.
Hopefully that makes sense.
@@JumpIntoAI What kind of LoRA? Do you use a specific one?
@@hmm-m Not one you can download. I mean a LoRA you have to create of that person.
I haven't shown how to make one because it can be a complicated process to get right.
Hello, I did all the steps using the image of a t-shirt, but when I go to generate the final image in Fooocus it only shows the t-shirt as if it were floating, it does not show the person's body. This happens with the last method you explain using 'Enable Mask Upload'
So you have an image of someone wearing a shirt and you are trying to change everything but the shirt right?
Do you have it setup just like mine where the original image is on the left and on the right is a white background image with a t-shirt blacked out? Both these images should be the same resolution and the right image (mask) needs a white background and not just transparent
If all that is true. What exactly is generating. Is everything around the t-shirt changing? just not turning into a person? Is the shirt changing and nothing else?
Is invert mask accidentally checked?
@@JumpIntoAI Hello, thanks for answering. The image is just the t-shirt, not of anyone wearing the t-shirt. I'm doing the steps you did with the image of the white dress with embroidered details, where you indicate that in the prompt: 'a woman modeling on the beach...'
Yes, I have the original image on the left and the black t-shirt on the right and they are the same resolution. The prompt I'm using is: 'a girl wearing a t-shirt at a music festival'
I generated 3 images, in 2 of them only the t-shirt with the music festival in the background appears but the girl does not appear. In image 3, the girl's shirt and arms appear, but the other parts of the girl's body, such as her head, neck, and legs, do not appear.
I already checked and yes, Enable Mask Upload is checked and Invert Mask is unchecked
@@ljimenez29 Ok, now i understand, this is just an image of the clothing item like my example of the dress.
This only works if the clothing item is in the shape of someone already wearing it, so if the t-shirt is just an image of it laid out or on a hanger then it wont work well.
Also if the item is the majority of the image and there is no room for a person that will need to be changed. Meaning, you need to have the white area around the image larger, like the t-shirt is further away so there is room to generate a person in it.
@@JumpIntoAI Ok, I get it. Thank you so much
@@ljimenez29 no problem, also forgot to mention a lot of the images with just the clothing item are on mannequins (which can work) but may also need to remove parts that you wouldn't see if someone wearing it like around the neck area like this
Before: imgur.com/a/f16pyZw
After: imgur.com/a/SFXx5bY
congratulations for the tutorial, but how and where to install it
"mashb1t" ?-thanks
github.com/mashb1t/Fooocus
Install it like you would fooocus just in a separate folder (you can change the config.txt to point at your model folders so you dont have to redownload them)
@@JumpIntoAI Thanks for the reply, I'm not familiar with it, could you please advise me or indicate a tutorial? Thank you
@@nicdisan To be found in the channel.
I have tried making photos via using the first method. But it didn't produce a photo with a human. It gave me my uploaded photo again.
The clothing item needs to be in the shape as if someone is already wearing it (such as on a mannequin). Also if this is a photo of an item where you can see parts you shouldn't if someone was wearing it (like a t-shirt but you can see the back of the neck/tag area) you will need to erase/cut that part out before making a black mask.
Example:
before: i.imgur.com/TY4X88j.png
after: i.imgur.com/pzDd3z7.png
Which GPU are you using?
4070 Ti 12gb
(I speed up the generations in the video to save time)
my generations are around 12seconds per image.
@@JumpIntoAI 12 secs is not bad at all.
It's stuck with the first prompt. It's taking forever and the process bar isn't moving :(
Is this the free colab version or installed on your computer? If free colab then this method probably uses too much vram for the free colab.
@@JumpIntoAI i am not sure. I just followed the steps. Is there a paid.install version?
Please make a video where you can install this mask in more detail.
I will do a video on the fooocus Mashb1t fork since I use that one over the Original one anyway.
@@JumpIntoAI thank you very much!!!
You are good looking guy
😊
Is there any way to change a person's skin tone using artificial intelligence?
you can certainly control the skin tone generated by a model (to a point) if specifying details such as the skin color, ethnicity, country.
If you are "only" trying to change the skin tone of an already generated image without altering its characteristics, then no. not without altering the model.
lllyasviel is the one who make fooocus
Very aware of that and I'm grateful for his work on fooocus, controlent, forge, etc. Certainly not taking away from that.
But he has stepped back from fooocus to work on forge and other projects for now and Mashb1t has been the main collaborator on the Dev page answering all my and everyone else's questions for months now. Just wanted to throw a little credit his way.