Training HD-22 with LoRA/Dreambooth and contributing to project

[hd-account]

I want to understand how this works more, and hopefully be able to contribute to the project, so I'm trying to start training the model. I'm following a guide, but I am not sure of some specifics so was hoping you could help me out.

• Is HD-22 based of stable diffusion 1 or 2? Do I train on 512x512 or 768x768?
• Should the negative prompt be included in the training?
• Is it best to merge the resulting model back with HD-22, or just use the newly trained model outright?
• If i were to train a model that includes a character that isn't currently available in HD, how could I contribute this to HD?

I have more than a decade of software engineering experience, if there's anything I can help with outside of training (even with the website) please let me know. HD has provided me with a lot of entertainment, so i really want to contribute.

Thanks!