No LoRA NEEDED! PuLID Same Character Creation Without LoRA with FLUX 1. GGUF - ComfyUI Tutorial

[@TheLocalLab]

🔴 Created a New Fluxgym Runpod Template for Training Flux Loras Faster at Low Cost 👉 kzbin.info/www/bejne/mmq9qqmuj6qefrs 👉 Want to reach out? Join my Discord by clicking here - discord.gg/5hmB4N4JFc

[@johnnyapbeats]

is it close to Kohya ? In terms of quality

[@TheLocalLab]

​@@johnnyapbeats Fluxgym actually uses the Kohya ss - sd3 branch - flux scripts as its backend. The repo itself mainly provides a user friendly interface with a few added on features making it easier to train the scripts.

[@johnnyapbeats]

@@TheLocalLab can I use Flux Gym on Kaggle is there a notebook available? Cause when I try to set everything up I get version mismatching

[@TomiTom1234]

Finally got it installed and working, thank you!

[@TheLocalLab]

No problem. I'm glad you made it through the tracebacks. Have fun with it.

[@RonnieMirands]

I gave up on this. After install correctly PullID using Comfyui Manager, it insists that need TRY TO FIX (cause it failed on Import) So it stays on this loop and nothing more than this :(

[@RonnieMirands]

I even install a new Comfyui/Manager fresh new installation and didnt work as well :(

[@SanchezGodsent]

@@RonnieMirands the same

[@sasansa9684]

i have error in sampling advance node, how can i fix it?

[@kalabushkin]

none of the bloggers say that the overall quality of generation when working with Pulid is much worse than Flux, yes, you will transfer the face, but the overall prompt will be made at a low level 😏

[@iCuteJP]

Cannot get the Pulid node install. I use the commands in the description but still not working. When it start it get " ImportError: DLL load failed while importing onnx_cpp2py_export:"

[@TheLocalLab]

yeah I came across several errors when trying to install the node my first time but didn't run into that one. You can try reinstalling on a fresh comfyUI install instead maybe.

[@francsharma7276]

can we change the face expression

[@TheLocalLab]

You can trying prompting a different facial expression in the positive prompt, like a smile or sadness and it could work. I was able to get generations not shown in the video of my character smiling pretty easily.

[@laurencecatalano2186]

it looks like a great workflow but... I can't get PuLID although I've tried everything I can think of

[@TheLocalLab]

Yeah this pulid node was a pain to install the first time. Did you try the commands in the description?

[@cgdtb]

Does Flux have an image-to-video model??

[@Aerisgel]

I got this error while running the workflow: RuntimeError: expected scalar type Half but found BFloat16 Any ideas to fix it?

[@TheLocalLab]

Did you load any models other then the default models from the workflow? Also is your Comfyui and custom nodes up to date?

[@Aerisgel]

@@TheLocalLab yes I loaded the models from the workflow and they are uptodate I think

[@TheLocalLab]

First check to make sure your using the properly "vae" model in the load vae node, reselect it if needed. If that does nothing, join the discord and provide a snapshot of the workflow you ran when you received the error and the error in the help channel. I'll take a closer look. Discord - discord.gg/5hmB4N4JFc

[@ovideotube]

great tutorial, it works alo with batch image as input? i mean with multiple pictures of the same person

[@TheLocalLab]

Currently the workflow is only set to work with one input image but can produce as many output images as you like.

[@renascienza.bazarclub]

This doesn't seem like character consistency, looks more like face switch, without a previous face.

[@AgustinCaniglia1992]

isn't that just FLUX 1? not 1.1 as described in the title of the video.

[@TheLocalLab]

Thank you for pointing that out. Issue fixed.

[@JhonEscobar-l5f]

please help, itts impossible for me to overcome the "`newbyteorder` was removed from the ndarray class in NumPy 2.0. Use `arr.view(arr.dtype.newbyteorder(order))` instead."

[@TheLocalLab]

You might need to use a different version of Numpy, maybe an older version.

[@JhonEscobar-l5f]

@@TheLocalLab 'GGUFReader' object has no attribute '_get_dtype' what this means?