Hello, I am narugo1992, the account holder, and leader of the DeepGHS team.
Since the buzz system was introduced on civitai, we have received numerous tips from different users. Many of these tips expressed recognition and appreciation for our team's work or conveyed the hope that we could train models for specific characters.
Yes, we have seen all of these messages, and they are deeply appreciated. Due to the current absence of a buzz-based messaging feature on civitai, we are posting this announcement to say, sincerely—thank you very much!!!❤️
From the moment we started writing the first line of code for training automation, we have been continuously striving to improve model quality and significantly reduce the manual effort required for training.
However, along the way, we have faced much criticism. On one hand, it was because the initial versions of the models indeed used low-quality base models to generate previews, giving the impression of low quality. On the other hand, some LoRA trainers who insist on manually training models and take extreme pride in it also held negative views of our automated workflow. We have endured a lot of negativity and criticism, most of which had nothing to do with our work or model quality—just because our workflow is fully automated! NO HUMAN's LOVE! THEY LACK SOUL! THEY ARE OFFENCING THE CHARACTERS! Any words you can imagine... The only things that have kept us going in this research are our own interest and the conviction of the correctness of the research direction of automated processes. We have been facing these criticisms, much like how carriage drivers were passionate about smashing steam locomotives during the Industrial Revolution.
However, it is you who have made me and my team realize that there are many people who have a positive attitude towards our work. In fact, many people like our models, and many people appreciate our research efforts. This unprecedented encouragement and confidence you have given us are truly invaluable.
In response to all of you, we will continue our research, and we have noted down all character LoRA requests mentioned in the tips. When the conditions are right, we will fulfill all of them.
BTW: We are currently undergoing a comprehensive technical upgrade of our automated training pipeline, and we will soon launch version 1.5 of the training automation process. However, our pipeline is still in a paused state for the time being, so for new LoRA requests, we can only process the data for now and will schedule training once the new version of the pipeline is online. Thank you for your understanding.