How to prepare data for controlnet training

[ymlab]

Is there a detailed description?

[kadirnar]

doc: https://github.com/XLabs-AI/x-flux?tab=readme-ov-file#training-dataset

Example Dataset: https://huggingface.co/datasets/kadirnar/fluxdev_controlnet_16k

[ymlab]

@kadirnar Hi, have you successfully trained the controlnet for flux?

[kadirnar]

@kadirnar Hi, have you successfully trained the controlnet for flux?

yes

[mashijie1028]

@kadirnar Hi! Thanks for your answer. Could you please provide some demo codes about the data loader of this example dataset https://huggingface.co/datasets/kadirnar/fluxdev_controlnet_16k?

[mashijie1028]

Oh! I figured it out!

Just the one-line code:

from datasets import load_dataset
ds = load_dataset("kadirnar/fluxdev_controlnet_16k")

[Oguzhanercan]

@kadirnar Hi, have you successfully trained the controlnet for flux?

yes

Hi, how many samples did you use to train a controlnet and can you mention about your compute-time resources?

[kadirnar]

@kadirnar Hi, have you successfully trained the controlnet for flux?

yes

Hi, how many samples did you use to train a controlnet and can you mention about your compute-time resources?

Yes, I trained it. However, it doesn't support multi-GPU. Therefore, I couldn't train it with a large dataset. That's why the results are not good.

[Oguzhanercan]

@kadirnar can you mention about the number of samples and compute device - time which fails in your case, I want to compare it with mine and estimate if I can scale it or not.

[kadirnar]

@kadirnar can you mention about the number of samples and compute device - time which fails in your case, I want to compare it with mine and estimate if I can scale it or not.

• 1xA100 and 17k image.

I think they have added multi-GPU support to the diffusers library.

[Oguzhanercan]

Do you remember how many iteration did you train it :)