I bought your AI course, I already know how to use ComfyUI but you definitely know all the nice tricks which I learned thanks to your videos, my interest in Flux is solely for the purpose of generating consistent AI girls and you are definitely the best in this subject. I really want to thank you for the video you made on how to train a Lora in Flux, I finally managed to get a good Lora. You are a true man of culture, keep it up.
Thanks for the work on this, I do believe plastic looking skin is a significant downside to Flux.1-dev. I've tried many ideas to overcome it. Most have failed to give me the skin I want AND maintain consistency in my character. Seems I can have one or the other. I'll give this shot. Subscribed and Liked.
itd be cool if u could show how to do all this on google colab and give estimates on time it takes to gen images there, for those without good comps. would make your course more accessible
If you add --lowvram and --disable-smart-memory to your bat file you can run proper dev without a quant version on your 3090. This also speeds it up because GGUF's are slower.
@@amxaas4450 Because Flux workflows with the normal models take up a lot of vram, so if you want to use them and not sacrifice any quality, switching to lowvram allows you to run the workflow without going oom, because it unloads models automatically.
Why not just use Unipc bh2 sampler with beta scheduler? you won't have to do anything else. Also, this method won't flark up your character if it's from a flux Lora. ;)
I want to ask if these clips on youtube you shared are different from the clips on your website? I want to register as a member on your website but I don't know if there is a difference between the clips on youtube and the private videos on the website?
this is a good method for close-up photos but Ive noticed for full body images it doesnt work very well due to the faceparsing either not being able to find the face or the face being too low resolution for it
@@NotHappy_Phantom it's in clip node, at least the one i'm using, you have guidance setting at the bottom, using 2 makes less plastic skin usually, the default is 3.5 or 4. You can go even lower but you'll see some abominations there
Could you please put the models on google drive or something. Baidu netdisk keeps trying to install an .exe file. I'm not going to do that....I don't imagine many Westerners would. Thanks.
LayerMask: PersonMaskUltra V2 [Errno 13] Permission denied: 'E:\\ComfyUI_windows_portable\\ComfyUI\\models\\mediapipe\\selfie_multiclass_256x256.tflite' Help me man there is a problem in your setup
@@RIP3Dtut yup just go to the "mediapipe" folder in models and copy the model in the mediapipe. Then delete the whole mediapipe folder and make a new folder in the same place and name it "mediapipe" and paste the file in inside it. But most probably you will one more error (a different one). just copy the error name and paste on Google then go to git and you will find the solution
Is it possible to fix fake looking Ai skin by bringing image created by Flux but outside of this workflow? How would I update the workflow to allow this?
🚀Last Change to get 40% Off of my Ai Model Course for beginners use code: AICONOMIST40 at checkout.
rebrand.ly/AI-Influencer-Model-Course
Hi 👋
I bought your AI course, I already know how to use ComfyUI but you definitely know all the nice tricks which I learned thanks to your videos, my interest in Flux is solely for the purpose of generating consistent AI girls and you are definitely the best in this subject.
I really want to thank you for the video you made on how to train a Lora in Flux, I finally managed to get a good Lora. You are a true man of culture, keep it up.
Thank you for your kind words, this truly means a lot! :)
Hi. How did you get the image from the flux model in the first part into the nodes of the second part?
Thanks for the work on this, I do believe plastic looking skin is a significant downside to Flux.1-dev. I've tried many ideas to overcome it. Most have failed to give me the skin I want AND maintain consistency in my character. Seems I can have one or the other. I'll give this shot. Subscribed and Liked.
If you add some kind of filter in another app you can avoid this. Most social images have a filter or a grain. Quick hack.
itd be cool if u could show how to do all this on google colab and give estimates on time it takes to gen images there, for those without good comps. would make your course more accessible
Thanks for the tutorial it was easy to follow along and build the work flow out as well.
Glad it helped!
If you add --lowvram and --disable-smart-memory to your bat file you can run proper dev without a quant version on your 3090. This also speeds it up because GGUF's are slower.
Why would you add the "loweram" prompt on a 3090 with 24GB?
@@amxaas4450 Because Flux workflows with the normal models take up a lot of vram, so if you want to use them and not sacrifice any quality, switching to lowvram allows you to run the workflow without going oom, because it unloads models automatically.
Why not just use Unipc bh2 sampler with beta scheduler? you won't have to do anything else.
Also, this method won't flark up your character if it's from a flux Lora. ;)
Sorry, im a newbie, what a Unipc bh2 sampler with beta scheduler? Thanks!
DISCLAIMER: Only works for half body photos, does not work for full body images
Thanks all the same @aiconomist
im using noice injection, its doing magic for skin
The more_details model is for SD1.5 and you're using an SDXL checkpoint. Hmmm
Can you use an already created image as input to basically detail and fix the skin???
Yes, this will also work
I want to ask if these clips on youtube you shared are different from the clips on your website? I want to register as a member on your website but I don't know if there is a difference between the clips on youtube and the private videos on the website?
hi, where do I find the specific download for the workflow shown in the video?
You can download workflows for members only on the website.
this is a good method for close-up photos but Ive noticed for full body images it doesnt work very well due to the faceparsing either not being able to find the face or the face being too low resolution for it
Contact me via email and i'll send you the workflow to solve this issue!
@@Aiconomist I emailed you :)
But you input the SDXL-Image into the FaceParser... shouldn't it be the Flux output that goes into the FaceParser? I am confused
How to do this for full body shots? I've tried it with those but it didnt work
Hello and thanks, where I can find that node "no negative prompt"?
It's just a clip text encode node that has been renamed.
Just use clip guidance between 2 and 2.4 max and you don't have this plastic faces anymore
yo bro can you tell me what is clip guidance and where i can change it? (i mean which node)
@@NotHappy_Phantom it's in clip node, at least the one i'm using, you have guidance setting at the bottom, using 2 makes less plastic skin usually, the default is 3.5 or 4. You can go even lower but you'll see some abominations there
Could you please put the models on google drive or something. Baidu netdisk keeps trying to install an .exe file. I'm not going to do that....I don't imagine many Westerners would. Thanks.
Need your help bro. Am using lightening ai cloud and the layerstyle is giving me error saying missing model
Facing same issue. Did you fix it?
hellow i cant use de LayerMask of lyerstyle i have got some errors
Where can I download the workflow??
You can download workflows for members only on the website.
LayerMask: PersonMaskUltra V2
[Errno 13] Permission denied: 'E:\\ComfyUI_windows_portable\\ComfyUI\\models\\mediapipe\\selfie_multiclass_256x256.tflite'
Help me man there is a problem in your setup
I have the same error, you got this fixed?
@@RIP3Dtut yup just go to the "mediapipe" folder in models and copy the model in the mediapipe. Then delete the whole mediapipe folder and make a new folder in the same place and name it "mediapipe" and paste the file in inside it. But most probably you will one more error (a different one). just copy the error name and paste on Google then go to git and you will find the solution
Is it possible to fix fake looking Ai skin by bringing image created by Flux but outside of this workflow? How would I update the workflow to allow this?
thanks :)
"Where can I download the workflow? Thanks."
по платной подписке иначе собирай сам по видео.😂😂😂
You can download workflows for members only on the website.
Please tell us about mixing LoRa in one stream (Mix LoRa)
These AI voices drive me crazy 🤬🤬🤬 Use your own voice!
[Errno 13] Permission denied: 'Z:\\Projects\\3rdparty\\comfyui\\models\\mediapipe\\selfie_multiclass_256x256.tflite' CAN ANYONE HELP WITH THIS!!
i am getting the same error. you got it fixed?