วีดีโอ

I don't have this error after following the steps

Been busy writing some comfyui nodes for myself that I am publishing in ComfyUI Manager search for WWAA. I did a test with Florence and llama3 vision but I found Florence descriptions are better for training loras. Llama3 was very restrictive and would set of its NSFW sensor even when image had women in bikinis. Stick with Florence - I'm going to share an updated and simplified workflow

Load the PNG into ComfyUI. Drag and drop

JSON workflow is embedded in the PNG

@ tks bro, 1 more problem. i click install the missing node but the apply ipadapter style is still missing, sorry for asking too much, i'm newbie

Set to true and then run ComfyUI. It will go back to 0 and start. After you run it once set it to false before you run it a second time.

You are right. You don't need to run the whole thing. Models will already be on persistent disk. However the ComfyUI dependencies would be wiped clean when you shut down the runpod as they are on non-persistent disk, so I suggest you run step 1 & 2 again. Nothing bad will happen but the dependencies will be installed so you can run ComfyUI again.

@@weirdwonderfulaiart Thanks for clarifying!

No I'm just a geek!

Did you try all the different steps? How about if you add the node again?

@@weirdwonderfulaiart yes , but turns out , i am a massive dingus , i was missing controlnets in the workflow that someone kindly made for me , that person had controlnets , now it works

Yep the custom node supports png only

No I don't think so.

@@weirdwonderfulaiart Thank you for the response. I guess with my setup that was the issue. I had to scale this huge image down to 1024 and it went through quick; the rest were normal sizes. Thank you for the tutorial!

Its available in the zip file via the blogpost link. Check in case your computer maybe quarantined the file ending with .IPYNB

First step in Environment Variables. There is a line where "Token" should be supplied. Update the value and run the cell before downloading Flux

Flux needs at least 12GB to my knowledge. Even if it runs it will take forever to produce anything

Most Welcome 🙏!! Thanks for the compliment.

try to run without Controlnet. You must have the set to True.

Awesome

Thanks for your feedback. Yes indeed ChatGPT can do that but I didn't want to complicate things. I have a ComfyUI workflow that lets you create the captions for free.

models\ESRGAN

Did you do all the suggested changes? Looks like your seed is not correct. Change the seed in high res fix node to 0

Yes I did. Thanks for picking the typo Storage is another great option but I know that people may not want to pay for ongoing volume storage.

That's great. Glad you fixed the incompatibility

Most Welcome 🙏!! Thanks for the compliment as well.

Most welcome!

Yes I can. Screenshot of the workflow is available here: weirdwonderfulai.art/resources/how-to-fix-the-high-res-fix-script/

Can you point to the timestamp of the video to which your questions relate to?

@@weirdwonderfulaiart the runpod template creates a jupyter notebook, the page where comfyui will display, a file uploader, and a code server. im wondering when u actually use the latter 2? i tried uploading a file once and it had an estimated time of hours and hours. way faster to do wget or aria2c. so just wondering use cases for either of those parts of the template

@@weirdwonderfulaiart 1:26 it shows all the ports made. im not seeing the use cases for the runpod file uploader, the application manager, or the code server.

Not sure I follow your question

However it's free and no need for paying subscription. Slowness may be due to your gpu.

@@weirdwonderfulaiart no, my gpu is doing very well. It's known that sdultimate upscaler is really slow.

Not really but thanks for the compliment

Thanks for pointing this out - the link is fixed now

@@weirdwonderfulaiart sorry it still does not work, it seems to be a direct download link; guessing TH-cam doesn't support it with their weird redirects

@@weirdwonderfulaiart a good option would be to put it in your github

I publish all my workflows via the blog. Still quite new to git world. I am sure the link is working now, i've tested it a few times.

Glad you think so. It is great way for me to batch stuff in ComfyUI

Indeed great point, captions are the key so the better defined they are the better LoRA you get.

Works just fine with the latest version of ComfyUI. See for yourself screenshot from today - weirdwonderfulai.art/wp-content/uploads/2024/12/Screenshot-2024-12-28-210257-scaled.jpg

I'll try it again with a fresh install. Perhaps I had some sort of conflict while I was testing last night. Thanks. EDIT: I tried a fresh install and it all now works perfectly. Perhaps there was a package install error last night that I didn't catch. I was actually looking to build my own workflow for this very reason when I came searching last night so thank you VERY much as you've saved me a lot of time!

@@truthwillout1980 sure let me know if you need any help

fp16 would give a better quality result than fp8

you can use fp16 if you have 24gb ram Runpod

yes, it should be saved with the prompt. You can also open the PNG in a notepad reader to see the metadata...search for seed.

@@weirdwonderfulaiart I finally figured out how to find all this info with forgeai, thank you for replying though!

I used this to create my captions and once finished I personally reviewed and adjusted them to ensure correct alignment with the image th-cam.com/video/5WiwquvlyHM/w-d-xo.html

Yes it still works but there are other methods available so it's up to you

Yes, it is. Download available here weirdwonderfulai.art/resources/cheap-solution-for-running-flux-dev-using-runpod/

Yes you can by modifying the notebook or using terminal with wget command

@@weirdwonderfulaiart I am trying to replicate my local comfy setting on runpod using network storage, over at workplace folder. Do you know if there's an efficient way to copy and paste my comfy folders (models, custom, Loras)? The dragging and dropping of folders doesn't seem to work.

This is what you get when you use this workflow.

Probably you are having too many training images and or Image Batch repeat is too high. Try by tweaking the parameters as shown in the video. I assume that you are on 24gb VRAM. If lesser then this could be the issue.

I will have a look at it.

@weirdwonderfulaiart thank u

Most welcome

Duplicate the load Lora node save then link it into the chain

@ sorry but i dont get it how to connect the node. Can you give me some examples?

It's very handy to have and monitor your resources

Thanks for your feedback

Works just fine even with the latest version of ComfyUI. See for yourself screenshot from today - weirdwonderfulai.art/wp-content/uploads/2024/12/Screenshot-2024-12-28-210257-scaled.jpg

No this cannot but there would be others available that you could do this with. Just need to explore the web