What about more than one character in the single image. If I want to use 3 or 4 lora trained actual models in a single frame. Any examples or how will I prompt.?
When you use a specific LORA, the AI generates any face in the image using that same LORA. The solution is to first create the image using the LORA, and once you're satisfied with the result, lock the seed of that image. Then, generate the image again without the LORA. Afterward, import both images into Photoshop, layer them on top of each other, and reveal only the face from the image with your face, hiding the rest of that layer, or you can use inpainting with flux and inpaint faces with or without Lora.
@@KK47.. go to your Hugging face account click on your profile picture and then your id. Here you can see your models. click on each one you want to download, then click open Files and versions and you can see the file about 172 MB. download it it and put it in your loras folder
thank you so much for the detail instructions, I followed them, but for some reasons, the imagine generate faces are out of the portion. anyway can fix it?? Thanks
@@musicterritory4678 your welcome. The prompt and weight of lora is important. Loar weight should be around 0.8 to 1. In 90% I set 1 for lora weight. Also play with guidance amount between 2.5 to 5
Yes you can. Just be sure to give it images that have body. Whatever you give to ai it will make the lora with them even the shape of fingers and nails and everything.
LoRA (Low-Rank Adaptation) training is a technique that allows you to fine-tune large AI models, like image generators, more efficiently and with less data. Rather than retraining the entire model from scratch, LoRA training adapts specific layers in a low-rank format, making the process faster, cheaper, and less resource-intensive. This method is particularly useful for personalizing AI outputs to match specific styles or tasks without extensive computing power
@Jockerai that was really useful thank you 🙏🙏 I have another question if you don't mind! I need to make models for my fashion brand and I need to reuse them for different outfits and poses, is that possible with flux and comfy UI ? And is it the best choice to make models for a fashion brand?
@@frstgear Actually, it’s better if you don’t do that. It’s better for the photos to have different sizes, with a mix of vertical and horizontal shots. The more variety in the photos-whether it’s color, lighting, location, outfits, or poses-the better your training will be.
I get this error after the model is trained, and the files do not populate my hugging face model. Error uploading to Hugging Face: 'ascii' codec can't decode byte 0xf0 in position 545: ordinal not in range(128)
But how can i make myself in different art styles? Like Anime , Illustration , 80's , futuristic etc. I don't want just to make realistic stuff! I've tried combining loras all together but the model just won't make my face good. Any help?
@@johnnyapbeats simply add proper words like "anime style, van Gogh style etc" in your prompt. If you are not good in English just ask for it from chatGPT in your language. Ask it to generate a good prompt for Fluxai like this: " write me a prompt for Fluxai a photo of me in anime style".
yes just click on your profile image and select setting, and then on the left you can see billing option click on that and put your billing information and charge your account
I want to create images using my face as a male character with a random girl or other characters standing beside me. However, when I use my trained model, the other characters in the image look similar to my face. For example, the model creates a female version of me as a girl. Is it possible to create different people in one image and have their faces look different from each other? For example, if I am standing next to Hugh Jackman, the model creates my face nicely, but Hugh Jackman's face also looks like mine.
You're right. When you use a specific LORA, the AI generates any face in the image using that same LORA. The solution is to first create the image using the LORA, and once you're satisfied with the result, lock the seed of that image. Then, generate the image again without the LORA. Afterward, import both images into Photoshop, layer them on top of each other, and reveal only the face from the image with your face, hiding the rest of that layer, or you can use inpainting with flux and inpaint faces with or without Lora
@@karansonkar2434 sometimes it can be frustrating but you see without ai or specially Flux some works were not doable. Just enjoy your work and make money, it can be satisfying.
on my ends it said 'training failed , Command '['python', 'run.py', 'config/replicate.yml']' returned non-zero exit status 1.' when i creat the training any help please ?
@@amirrezavahedi-bf7fh yes it's because you are renting a 2200$ GPU and it is very fast. Otherwise you can train this lora on your own system. If you have 4090 24G VRam GPU it may take 5 hours to train a lora locally
@@faketejagaming9705 actually no there isn't. Because Flux.pro is a very expensive model and they offered the free version to attract more people to their website. however at this time you have some free credits but it's limited. you will given free credit every 6 hours
![ละลาย (LALALYE) - DAOU PITTAYA ต้าห์อู๋ พิทยา [OFFICIAL MV]](http://i.ytimg.com/vi/wo6r9TgausI/mqdefault.jpg)
click the link below to stay update with latest tutorial about Ai 👇🏻 :
www.youtube.com/@Jockerai?sub_confirmation=1
Professional and clear content, Nice dadash
@@gammastudio3126 thank you bro. How did you find this video?
Just 6700 steps and voila - you’re done!
you made it by 6700 steps???
means? it bricked or made it super Realistic or what?
What about more than one character in the single image. If I want to use 3 or 4 lora trained actual models in a single frame. Any examples or how will I prompt.?
When you use a specific LORA, the AI generates any face in the image using that same LORA. The solution is to first create the image using the LORA, and once you're satisfied with the result, lock the seed of that image. Then, generate the image again without the LORA. Afterward, import both images into Photoshop, layer them on top of each other, and reveal only the face from the image with your face, hiding the rest of that layer, or you can use inpainting with flux and inpaint faces with or without Lora.
@@Jockerai thanks 😊 will try this
So if I download the ready-made model into the ComfyUI\models\loras\ folder, can I generate images localy using the provided keyword?
@@filmezes yes of course you can.
@@Jockerai can you show us how to download the lora from replicate then use in comfyui, I cant find anything on how to download
@@KK47.. go to your Hugging face account click on your profile picture and then your id. Here you can see your models. click on each one you want to download, then click open Files and versions and you can see the file about 172 MB. download it it and put it in your loras folder
could we train a lora in my local computer ?
my GPU Is GTX 1070Ti 8 VRam
you can but it takes very long time and in my opinion it's not safe and you hardware might be at risk. For your GPU it may take more than 24h
Error(s) in loading state_dict for DoubleStreamBlockLoraProcessor:
Missing key(s) in state_dict: "qkv_lora1.down.weight", "qkv_lora1.up.weight", "proj_lora1.down.weight", "proj_lora1.up.weight", "qkv_lora2.down.weight", "qkv_lora2.up.weight", "proj_lora2.down.weight", "proj_lora2.up.weight".
thank you so much for the detail instructions, I followed them, but for some reasons, the imagine generate faces are out of the portion. anyway can fix it?? Thanks
@@musicterritory4678 your welcome. The prompt and weight of lora is important. Loar weight should be around 0.8 to 1. In 90% I set 1 for lora weight.
Also play with guidance amount between 2.5 to 5
A month later now is there a process not to complex?
What do you mean?
I'm not worried about the face I want to create a lora that will give me a consistent body. Can I do that with this?
Yes you can. Just be sure to give it images that have body. Whatever you give to ai it will make the lora with them even the shape of fingers and nails and everything.
what is the simple definition of Lora training ?
LoRA (Low-Rank Adaptation) training is a technique that allows you to fine-tune large AI models, like image generators, more efficiently and with less data. Rather than retraining the entire model from scratch, LoRA training adapts specific layers in a low-rank format, making the process faster, cheaper, and less resource-intensive. This method is particularly useful for personalizing AI outputs to match specific styles or tasks without extensive computing power
@Jockerai that was really useful thank you 🙏🙏
I have another question if you don't mind!
I need to make models for my fashion brand and I need to reuse them for different outfits and poses, is that possible with flux and comfy UI ? And is it the best choice to make models for a fashion brand?
For the training images, do I need to resize them or put them in a certain aspect ratio?
@@frstgear Actually, it’s better if you don’t do that. It’s better for the photos to have different sizes, with a mix of vertical and horizontal shots. The more variety in the photos-whether it’s color, lighting, location, outfits, or poses-the better your training will be.
@@Jockerai thanks!
When i click create training it keeps saying the upload photos field is empty.
@@jonathanhilliard971 change your browser it will be solved
Nice! But I have a question, what of the person has tattoos? With SD Lora it was impossible to get good results. Have you tried so?
@@adrianmunevar654 no I didn't test any model with tattoos. Did you?
@@Jockerai I did yesterday and doesn't work :(
@@adrianmunevar654 I did couple of weeks ago and it worked. Send your prompt
I get this error after the model is trained, and the files do not populate my hugging face model.
Error uploading to Hugging Face: 'ascii' codec can't decode byte 0xf0 in position 545: ordinal not in range(128)
did you create your model in hugging face and create your token?
@@Jockerai yes
@@tysonwigley make sure your model on hugging face is public. If they are public I really don't know why this happen.
But how can i make myself in different art styles? Like Anime , Illustration , 80's , futuristic etc. I don't want just to make realistic stuff! I've tried combining loras all together but the model just won't make my face good. Any help?
@@johnnyapbeats simply add proper words like "anime style, van Gogh style etc" in your prompt. If you are not good in English just ask for it from chatGPT in your language. Ask it to generate a good prompt for Fluxai like this: " write me a prompt for Fluxai a photo of me in anime style".
@@Jockerai I've already done that but the model doesn't seem to grasp my features good , maybe I done something wrong on the training part ?
@@johnnyapbeats test it with lora weight between 0.87-0.95
@@Jockerai do you have any socials we can talk to in the future ?
@@johnnyapbeats yes search jocker.ai in Instagram you,ll find me.
should i add 2.5$ to replicate site? and how?
yes just click on your profile image and select setting, and then on the left you can see billing option click on that and put your billing information and charge your account
I want to create images using my face as a male character with a random girl or other characters standing beside me. However, when I use my trained model, the other characters in the image look similar to my face. For example, the model creates a female version of me as a girl. Is it possible to create different people in one image and have their faces look different from each other? For example, if I am standing next to Hugh Jackman, the model creates my face nicely, but Hugh Jackman's face also looks like mine.
You're right. When you use a specific LORA, the AI generates any face in the image using that same LORA. The solution is to first create the image using the LORA, and once you're satisfied with the result, lock the seed of that image. Then, generate the image again without the LORA. Afterward, import both images into Photoshop, layer them on top of each other, and reveal only the face from the image with your face, hiding the rest of that layer, or you can use inpainting with flux and inpaint faces with or without Lora
@@Jockerai I feel overwhelmed by the task of editing photos using Flux AI. Do you have any tutorial videos that you can suggest?
@@karansonkar2434 sometimes it can be frustrating but you see without ai or specially Flux some works were not doable. Just enjoy your work and make money, it can be satisfying.
@@Jockerai ok
on my ends it said 'training failed , Command '['python', 'run.py', 'config/replicate.yml']' returned non-zero exit status 1.' when i creat the training any help please ?
do it again sometimes you may face this error
are you currently living in Iran? how did you pay if you are
No I'm not why you thought that way?
@@Jockerai because of girls name :D
@@Averjax yes I'm familiar with them hamvatan 😁 dorost gofti
Nice work bro .. but ineed something free because need tested first ❤
@@DznStd the free version is to run this locally which needs at least 24G VRam for 5 hours computing. For lower it takes more. What GPU you have?
@@Jockerai think you bro 🌺
داداش دمت گرم ولی سایتی سراغ نداری که رایگان لورا رو بتونیم بسازیم؟
@@ActorClip-hl1me there is a way to train a lora locally I will make a video for that. Stay tuned
نایس
Thanks
LINK EXPIRED BROTHERRRRRRRRRRRRRR
@@ProductsDiaries ok bro I will fix in couple of minutes 🌱
all links are fresh now
@@Jockerai still same
"It looks like that invite has expired. Please ask whoever sent it to you to send a new one"
yes it is and it has reached it's limitation sorry about that.
Do I have to spend $2.50 every time I want to train loras at this site?
@@amirrezavahedi-bf7fh yes it's because you are renting a 2200$ GPU and it is very fast. Otherwise you can train this lora on your own system. If you have 4090 24G VRam GPU it may take 5 hours to train a lora locally
Is this paid or free??
@@faketejagaming9705 on replicate websites yes every model training costs about 2.5$
@@Jockerai is there any free way for that broo?
@@faketejagaming9705 actually no there isn't. Because Flux.pro is a very expensive model and they offered the free version to attract more people to their website. however at this time you have some free credits but it's limited. you will given free credit every 6 hours
th-cam.com/video/_rjto4ix3rA/w-d-xo.html&ab_channel=MattWolfe
But you get 10 dollar credit from the link of this video. It's enough for some Tests,right ?