Please please please make a video on training LoRA concept(s) with thousands of training images 😢 your tutorials are the best and most clear cut out there
@@AI-HowTo hands are still a bit hit & miss (though better, I guess) but faces in wide shots (or prompts more complex than "portrait of personX") are much better in like 99 % of generations.
true, I think i got better results for hands with a Lycoris from civitai before, i belive it was this one civitai.com/models/47085?modelVersionId=55199 but like you said, getting better hands is slightly difficult because of the direction of the fingers/hand in relation to the arm...etc, while face fix is very effective.
Great to see you again, im trying Locos from yesterdays videos today with great results. Id love to see an advanced adetailer video with the 2nd person option etc
Thanks, good, you should try with regularization and without, because i got contradicting results with that, some work better with class images some don't, so experimentation is very important, best of luck, will also check what you mentioned too when possible.
I have installed other extensions before but this one simply won't show up, it says it's installed but the tab doesn't show up and it doesnt even show up in the settings. I have restarted the terminal , checked for updates and even restarted the computer, running automatic1111 locally
I have not faced this, but I suggest you to delete the extension folder in stable-diffusion-webui\extensions\adetailer and install it again or use git pull to do update , hopefully that wil resolve anything missing in the installation in the first place
Another generous piece of knowledge from you, many thanks! I was going to investigate Adetailer extension already, but I didn't know it could cut out the img2img step to refine faces on character loras. Even some of the best trained loras on CivitAi absolutely need an extra rendering step to get close to the intended likeness. I wonder if in future Roop would also be part of the automatic process? And ControlNet Reference is also useful to help likeness.
true, After detailer seems very important in most LoRAs, autoamtic inpaint is really useful and saves time, despite having limited control over results, but it is better than Roop and gives finer results.
I don't remember right now what was the thing, but when i used ADetailer with meichdarkmix or such, I was getting really deformed faces. Original generations were just great on their own but when i tried testing Adetailer, it just lowered the overall image quality and deformed the faces. I am not sure what was wrong and I didn't bother with AD again. I will try again with Original ugly faces. I hope it might work better on models with bad faces as default.
after detailer should not be used if the faces are already good.... also, test the command to be used in after detailer first separately, to make sure that it produces good protraits, so that the automatic inpaint works well... in general, it is able to improve face quality in majority 90%+ of the images with lower quality or lower detailed faces, i think it is one of the most useful extensions
sorry not sure, you can double installed the !After detailer, not the other one without !, which can also be installed from URL github.com/Bing-su/adetailer .... it still works even in A1111 version 16.0 .... you may also consider using gitpull to update the installation to make sure you have most recent update ... or apply update and restart from the extension folder hopefully that works...otherwise not sure what is the problem.
Hey…nice video, very informative! Would like to know if you would be interested in a paid project and help us with some AI based image processing. Please share your email for us to send you the details.
Always great tutorials...will check this out.
Please please please make a video on training LoRA concept(s) with thousands of training images 😢 your tutorials are the best and most clear cut out there
will see if i can soon.
@@AI-HowTo hurray! 🤠
a small note: never ever show comparing pictures in order: after - before... is quite confusing 🤣
point taken, thanks
This is a complete game changer !! Tried with a few custom loras and the differences are staggering
indeed, this extension is a must have, should be installed automatically by A1111, can be used also with hands since these are major issues in SD.
@@AI-HowTo hands are still a bit hit & miss (though better, I guess) but faces in wide shots (or prompts more complex than "portrait of personX") are much better in like 99 % of generations.
true, I think i got better results for hands with a Lycoris from civitai before, i belive it was this one civitai.com/models/47085?modelVersionId=55199 but like you said, getting better hands is slightly difficult because of the direction of the fingers/hand in relation to the arm...etc, while face fix is very effective.
Great to see you again, im trying Locos from yesterdays videos today with great results. Id love to see an advanced adetailer video with the 2nd person option etc
Thanks, good, you should try with regularization and without, because i got contradicting results with that, some work better with class images some don't, so experimentation is very important, best of luck, will also check what you mentioned too when possible.
I have installed other extensions before but this one simply won't show up, it says it's installed but the tab doesn't show up and it doesnt even show up in the settings.
I have restarted the terminal , checked for updates and even restarted the computer, running automatic1111 locally
I have not faced this, but I suggest you to delete the extension folder in stable-diffusion-webui\extensions\adetailer and install it again or use git pull to do update , hopefully that wil resolve anything missing in the installation in the first place
Another generous piece of knowledge from you, many thanks! I was going to investigate Adetailer extension already, but I didn't know it could cut out the img2img step to refine faces on character loras. Even some of the best trained loras on CivitAi absolutely need an extra rendering step to get close to the intended likeness. I wonder if in future Roop would also be part of the automatic process? And ControlNet Reference is also useful to help likeness.
true, After detailer seems very important in most LoRAs, autoamtic inpaint is really useful and saves time, despite having limited control over results, but it is better than Roop and gives finer results.
Can't wait for sdxl 1.0 tutorials once it gets released
I don't remember right now what was the thing, but when i used ADetailer with meichdarkmix or such, I was getting really deformed faces. Original generations were just great on their own but when i tried testing Adetailer, it just lowered the overall image quality and deformed the faces. I am not sure what was wrong and I didn't bother with AD again. I will try again with Original ugly faces. I hope it might work better on models with bad faces as default.
after detailer should not be used if the faces are already good.... also, test the command to be used in after detailer first separately, to make sure that it produces good protraits, so that the automatic inpaint works well... in general, it is able to improve face quality in majority 90%+ of the images with lower quality or lower detailed faces, i think it is one of the most useful extensions
Can you make a video comparing lora with SD vs lora with SDXL ?
will do.
Is there a similar extension for this in ComfyUI?
not sure, but this extension is for A1111, it is not developed for Comfy UI.
hi dude i installed adetailer but not apear after restart
sorry not sure, you can double installed the !After detailer, not the other one without !, which can also be installed from URL github.com/Bing-su/adetailer .... it still works even in A1111 version 16.0 .... you may also consider using gitpull to update the installation to make sure you have most recent update ... or apply update and restart from the extension folder hopefully that works...otherwise not sure what is the problem.
support XYZ ??
Yes, it seems options for After detailer are available for this even in X/Y/Z prompt, in recent versions of A1111
Thanks!
You are Welcome!
Hey…nice video, very informative! Would like to know if you would be interested in a paid project and help us with some AI based image processing. Please share your email for us to send you the details.
Thank you so much for the offer, unfortunately currently I am not considering any new work, thank you once again.