Is one trainer the best way to train a lora?
Is one trainer the best way to train a lora?
Fboth kohya and onetrainer same
Fconfig matters
T@Dr. Furkan GözükaraI know you haven't done much in comfy, but I feel like this being more of a computer science question (and probably would or wouldn't work the same way for a1111) and you'd probably be pretty knowledgeable on the matter of venv.
I've been trying to update comfyui to the updated build to get with the times. Tried updating on my primary (leaving alone the legacy one I've been running because I want to be able to run it as a backup), but it ended up with an error message that kept both from running. Got spooked at the idea of just my shit never running again and did a restore point back up. Somehow, it worked and I could generate again.
Hoping to kill two birds with one stone here by 1) updating to current/see how I can get it all functioning properly and 2) learn proper venv use that I have been slacking on.
I get the feeling that using virtual environments for all/both of my builds is really what I need to be doing in order to see that I can keep the one working/troubleshoot the other as needed.
Is my assumption right about virtual environments correct? And, if so, how can I start to get one functioning with my legacy build? I.e. Would like to be able to launch it regardless of what I update. I'm guessing that updating the current one and trying to get it working through venv would product the same outcome for the legacy (that it's still working), but I also want to be able to ensure it can still function in case I find myself updating system stuff in a non venv way that would mess with the legacy.
Big sorries for any newbie things I'm getting wrong. lol.
I've been trying to update comfyui to the updated build to get with the times. Tried updating on my primary (leaving alone the legacy one I've been running because I want to be able to run it as a backup), but it ended up with an error message that kept both from running. Got spooked at the idea of just my shit never running again and did a restore point back up. Somehow, it worked and I could generate again.
Hoping to kill two birds with one stone here by 1) updating to current/see how I can get it all functioning properly and 2) learn proper venv use that I have been slacking on.
I get the feeling that using virtual environments for all/both of my builds is really what I need to be doing in order to see that I can keep the one working/troubleshoot the other as needed.
Is my assumption right about virtual environments correct? And, if so, how can I start to get one functioning with my legacy build? I.e. Would like to be able to launch it regardless of what I update. I'm guessing that updating the current one and trying to get it working through venv would product the same outcome for the legacy (that it's still working), but I also want to be able to ensure it can still function in case I find myself updating system stuff in a non venv way that would mess with the legacy.
Big sorries for any newbie things I'm getting wrong. lol.
Fyour first one was portrable?
Fi have comfyui installer on patreon
Fwhich installs into a venv
Fif you use that it will work individually without affecting each other
Fcomfyui portable is breaking entire system i seen on others
Fso if your legacy was manual install into venv it would keep working
Falso comfyui portable sometimes breaking your default python and it causes break in other apps
Fi am making installer for LivePortrait animal version
Dthere are multiple clothes changer/try-on apps for SD from what I can tell, are any of them clearly above the rest?
F
Fidm vton is best as far as i know
Frope pearl live test
F
Fworking perfect with tensorrt
N@Dr. Furkan Gözükara
NHello, Thank you very much for your availability and your help, I installed LivePortrait with the Kaggle notebook despite the weakness of my PC
Fyou are welcome
Frope live real time swap video recorded time to edit
PDoes anyone have a tutorial or a good article for using controlnet in SDXL?
I tried to use SDXL with controlnet but it doesn't work as good as with SD1.5.
Maybe I used wrong controlnet models
I tried to use SDXL with controlnet but it doesn't work as good as with SD1.5.
Maybe I used wrong controlnet models
N@Dr. Furkan Gözükara any chance for implementing https://github.com/plemeri/InSPyReNet or https://github.com/ZhengPeng7/BiRefNet for SD dataset image bg removal and/or masking?
GitHub
Official PyTorch implementation of Revisiting Image Pyramid Structure for High Resolution Salient Object Detection (ACCV 2022) - plemeri/InSPyReNet
GitHub
[CAAI AIR'24] Bilateral Reference for High-Resolution Dichotomous Image Segmentation - ZhengPeng7/BiRefNet
Fsure i can but i think they are not best
Fhave you tested them?
FRMBG is best as far as i know
Fthis repo says they are better
FGitHub
[CAAI AIR'24] Bilateral Reference for High-Resolution Dichotomous Image Segmentation - ZhengPeng7/BiRefNet
Fso this one looks promising
Fi am upgrading our app to cogvlm v2
Fdmn the models are like 30 40 gb
Many difference ?
Fdownloading models 40 GB 
F
Fbut i am pretty sure it will be best
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dwhat is it?
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
FCogVLM image captioning
FGitHub
GPT4V-level open-source multi-modal model based on Llama3-8B - THUDM/CogVLM2
MHey friends! I wan Wondering if someone can help me implement the IDM VTON model to make inference, I tried everything, I'm só frustrated
MI'm having a hard time with it
MIs it possible to run like without comfyui or gradio
MI struggle to find the right checkpoints and reference them to the commands
Fyou can use our gradio app source code easiest way
MHey doc, thanks for answering. But can I use as an api even using with gradio? Would it be possible to create an api for it
Fgradio gives you api
Fbut first you need to start gradio app
MOh nice, I'll try that way
MI found a step by step implementation too
MGitHub
After investigating for 3 days, I got almost all done except for some minor problems. Here is a link of personal made study case of HR-VITON Pre According to explanation from authors: Preprocessing...
MBut there's a lot of broken stuff