Already watched the whole video and I have a question: What if you create a workflow in which after generating the image, you can inpaint it, mask the face you want to replace, and in that inpaint section, you have another lora loader for selecting the one you want to use? Would it work? I'm really a noob in ComfyUI, so I don't know how to make it, but I'm sure you can. And I think that way it would be easier and faster. Don't you think so? Please let us know. 💪🏻🔥
@@adrianmunevar654 I tested many ways like inpainting, openpose, canny or other ones, but this method which I share in this video was the best one with most similarity to characters and most quality for images. Masking and inpainting with flux didn't work well for creating personal character (using personal loras). It fails a lot since flux doesn't reach that high quality in controlnets. Actually I prefer to use controlnet as less as I can in flux. Because It needs more improvements.
Hi, great tutorial, thank you, but I have a problem. I added the workflow and put the files exactly as in the video. I added my loras and when I click on generate image I get this window with the following message: SamplerCustomAdvanced module 'tensorflow' has no attribute 'Tensor' The process stops right in the sampler custom advance box
It can be done entirely in ComfyUI with Flux Inpainting. Create the base image of two people with LoRA of person A. Then use a Flux Inpainting workflow, mask out the person B's face. Use the same prompt and noise value, but the LoRA of person B. You could even automate it with face detection and automatic masking and then combine both workflows into one. This also works with basically any amount of LoRAs but I found 3 people is often the maximum where Flux can produce good results.
I have tested it via inpainting but the quality and matching faces are not as good as my method and you will get deform faces or something like that. But if you think it is possible share some photos please
@@Jockerai I just found out those problem with inpainting as well. But you can use a open pose or deep cnet to keep those faces from deforming while inpainting.
@@Paulo-ut1li Maybe you don't believe me but before I record a tutorial video, I test all of these ways that you are testing and then I share the best possible method. with open pose we have another problem that is the images has very low quality in colors and details, second the faces you generated are not similar enough to the person (personal lora) and you have to increase controlnet strength which make the image very low quality :) . So the best way for the best quality and best possible similarity to our person ( Lora) is the way which I explained in the video.
@Jockerai ai really believe, because photoshop is the software I end up using all the times for time saving, It's just more practical. Stable Diffusion is really bad at allowing for multiple characters Lora or composition. OmniGen and NVIDIA new models seems promising, let's see.
Is there any way to mask it on comfyui, because i would like the two characters placed together without the help of photoshop, and its a lot of work >.>
interesting way but I don't think it's gonna work because for outpainting which I taught in my previous videos we use SDXL models first that doesn't support flux Loras. :)
Hello. I am writing from Turkey. I've been searching for this for a long time. Thank you first of all. What I want to ask here is that I create 2 images by entering the same seed value. I create 2 images with 2 different character lores and the same promts and values. Even though I enter the same seed, it cannot do the exact same thing. My aim here is not only to put the face but also the anatomical structure of the man, along with his hair and beard, side by side. Did I ask for too much? hahahahah
A warm hello to everyone in Turkey from here! What you want to achieve is not possible by only keeping the seed value fixed because when the LoRA changes, the AI's input data becomes different. One way to approach this is to keep generating until you get a result that is approximately satisfying. Another method is to check out the tutorial "I Mastered Flux Redux" and use the image you’ve already generated as input. This way, you can create similar images with different LoRAs. This method will likely work well for your needs. Another way is to wait for the advanced inpainting tutorial that I will be uploading to the channel. You can use it to modify the second person's face. Make sure to subscribe and stay tuned!
@Jockerai There is a Tensor Art website, bro flux and you can train and use models on it. Now how can we combine 2 characters in 1 model with comfyui? For example, a woman next to a man.
Yes of course. You can put a prompt that have three person in it. First one with person A LoRA. second one person B LoRA and third image with person C LoRA. Then put every all images on top of eachother. Then mask faces you don't need
I tested many ways like inpaintingT openpose canny or otherones, but this method which I share in this video was the best one with most similarity to characters and most quality for images
Also it can be done for 3 or more personal loras guys, you just need to repeat the process :) . Don't forget to like the video guys❤️
I trust you my friend, just started watching your video and I already liked it because I know your method will work.
Already watched the whole video and I have a question:
What if you create a workflow in which after generating the image, you can inpaint it, mask the face you want to replace, and in that inpaint section, you have another lora loader for selecting the one you want to use? Would it work?
I'm really a noob in ComfyUI, so I don't know how to make it, but I'm sure you can. And I think that way it would be easier and faster. Don't you think so?
Please let us know. 💪🏻🔥
Thank you bro happy to hear that💚🙏
@@adrianmunevar654 I tested many ways like inpainting, openpose, canny or other ones, but this method which I share in this video was the best one with most similarity to characters and most quality for images. Masking and inpainting with flux didn't work well for creating personal character (using personal loras). It fails a lot since flux doesn't reach that high quality in controlnets. Actually I prefer to use controlnet as less as I can in flux. Because It needs more improvements.
Great video! I have been trying to do this for a bit, this helped.
@@VitoSansevero you're welcome bro 😉✨
Hi, great tutorial, thank you, but I have a problem. I added the workflow and put the files exactly as in the video. I added my loras and when I click on generate image I get this window with the following message: SamplerCustomAdvanced
module 'tensorflow' has no attribute 'Tensor'
The process stops right in the sampler custom advance box
It can be done entirely in ComfyUI with Flux Inpainting.
Create the base image of two people with LoRA of person A. Then use a Flux Inpainting workflow, mask out the person B's face. Use the same prompt and noise value, but the LoRA of person B.
You could even automate it with face detection and automatic masking and then combine both workflows into one. This also works with basically any amount of LoRAs but I found 3 people is often the maximum where Flux can produce good results.
I have tested it via inpainting but the quality and matching faces are not as good as my method and you will get deform faces or something like that. But if you think it is possible share some photos please
@@Jockerai I just found out those problem with inpainting as well. But you can use a open pose or deep cnet to keep those faces from deforming while inpainting.
@@Paulo-ut1li Maybe you don't believe me but before I record a tutorial video, I test all of these ways that you are testing and then I share the best possible method. with open pose we have another problem that is the images has very low quality in colors and details, second the faces you generated are not similar enough to the person (personal lora) and you have to increase controlnet strength which make the image very low quality :) . So the best way for the best quality and best possible similarity to our person ( Lora) is the way which I explained in the video.
@Jockerai ai really believe, because photoshop is the software I end up using all the times for time saving, It's just more practical. Stable Diffusion is really bad at allowing for multiple characters Lora or composition. OmniGen and NVIDIA new models seems promising, let's see.
@@Jockerai By the way, It would be awesome to see a video about the Flux Area Prompt and its capabilities.
Really cool!
thank you✨
Great content thank you bro!
Thanks a lot
Hi, if I don't know how to train lora but I use PULID for flux, can I use your method? Can you create a process for this idea? Thanks
You can make characters with Pulid too, and use my method to mask one of persons. But I personally didn't test Pulid for this
Is there any way to mask it on comfyui, because i would like the two characters placed together without the help of photoshop, and its a lot of work >.>
not bad. i use masking for 2 characters
Could you just generate with first Lora and then use outpaint for the second?
interesting way but I don't think it's gonna work because for outpainting which I taught in my previous videos we use SDXL models first that doesn't support flux Loras. :)
Hello, I have a 6gb vram system. Can I do the same? If it is the same, I want to buy it.
Buy what bro?
Hello. I am writing from Turkey. I've been searching for this for a long time. Thank you first of all. What I want to ask here is that I create 2 images by entering the same seed value. I create 2 images with 2 different character lores and the same promts and values. Even though I enter the same seed, it cannot do the exact same thing. My aim here is not only to put the face but also the anatomical structure of the man, along with his hair and beard, side by side. Did I ask for too much? hahahahah
A warm hello to everyone in Turkey from here! What you want to achieve is not possible by only keeping the seed value fixed because when the LoRA changes, the AI's input data becomes different. One way to approach this is to keep generating until you get a result that is approximately satisfying. Another method is to check out the tutorial "I Mastered Flux Redux" and use the image you’ve already generated as input. This way, you can create similar images with different LoRAs. This method will likely work well for your needs. Another way is to wait for the advanced inpainting tutorial that I will be uploading to the channel. You can use it to modify the second person's face. Make sure to subscribe and stay tuned!
Bro It would be better if you could upload this as flux for tensor art, Lora. Thanks in advance.
@@Victor_reis29 what do you mean I couldn't understand your comment!
What do you mean by tensor art?
@Jockerai There is a Tensor Art website, bro flux and you can train and use models on it. Now how can we combine 2 characters in 1 model with comfyui? For example, a woman next to a man.
@Victor_reis29 I have fully explained that in the video
@@Jockerai thank you bro I subscribed 🙏 😅
3 people possible, sir?
Yes of course. You can put a prompt that have three person in it. First one with person A LoRA. second one person B LoRA and third image with person C LoRA. Then put every all images on top of eachother. Then mask faces you don't need
@@Jockerai thx a lot, it seem that you tried already, test ended. right, sir? Anyway thx as always. I respect you, sir. : )
@@ismgroov4094 you're welcome my friend, yes I shared a short video which has an image with three loras at the same time in an image 😉✨
sorry i will inpaint with mask on second lora
Could you share the workflow please?
I tested many ways like inpaintingT openpose canny or otherones, but this method which I share in this video was the best one with most similarity to characters and most quality for images
How could you be the best teacher in teaching Ai specially in Flux😍. Best tutorials you have bro keep doing that I love your teaching way❤
thank you bro I really appreciate your kind words❤🤗