Future Plans

[Linaqruf]

I'm sorry for my slow response these past two weeks. There are several reasons for my delay:

• Ramadan plans took priority over my personal plans, leaving me with little time for anything else. Now that Eid Mubarak has passed, I have more free time.
• I'm growing tired and losing interest in maintaining this repo. There are simpler and better Colab notebooks available, and I thought I could finally take a break. Unfortunately, I didn't realize that there are people who still using this repo, I'm sorry
• Recently, I've become more interested in contributing to the main repo and developing Cagliostro Colab UI.
• Every day, new arguments and LoRA modules are released, and I'm struggling to simplify everything while also explaining all the parameters.

So for the future plans:

• I'm planning to discontinue the Kohya Dreambooth notebook after the last two weeks of Fast Kohya Trainer. I rarely use this notebook, and it has fewer visitors than other notebooks, so I can't maintain it anymore.
• I plan to remove all mentions of "multi-concept training." I feel that V15 was a bad update, and all the automatic recursive features don't align with my goal of creating a flexible notebook.
• I'm planning to restore the sample_prompt cell to the notebook.
• I'm planning to mirror Kohya Trainer to Huggingface, so all future updates will be published there before being released to this repo.
  • I'm planning to clone the main repo instead of mine since it's more up-to-date. This repo is only a fork of his, created before Kohya joined Github.
  • I'm planning to separate the dataset preprocessing section from the training section and create a dataset_annotator notebook.
  • Because it's on Huggingface, I want to use diffusers instead of compvis format (except LoRA).
  • To maintain the quality and reproducibility, I'll only release it to Github when it has fewer bugs.

Please give suggestion and feedback, thank you!

[Isaac-DR]

may i ask which are those notebooks? to they run on google colab?

There are simpler and better Colab notebooks available

[anothertal3]

Thank you for the information. Your intentions all seem very reasonable to me.

and it has fewer visitors than other notebooks, "Visitors" will probably only be counted if they access the notebook via your repo, I assume? I think there will be many who, like me, did a copy of the notebook in order to keep their settings between sessions. You're "visitor counts" might be way lower than the number of your real users ;)

(That being said I am only using the "Kohya LoRA Trainer" myself)

[anothertal3]

may i ask which are those notebooks? to they run on google colab?

There are simpler and better Colab notebooks available

@Isaac-DR Did you find any of those in the meantime, by chance?