Replies: 1 comment
-
SDXL LoRa models are trained on an XL checkpoints and then used with Turbo and Lighting. I'm not 100% sure, but I don't think you can successfully train on Turbo or Lightning Checkpoints given the nature of how they work. I could be wrong though. I've never seen, or heard of, a trainer that lets you specify the CFG Scale and Steps. Those are parameters specific to generating images, not training a model. Though some settings during training could dictate the tolerances of those parameters, but I can only see that being applicable to training Checkpoints and not LoRa. EDIT: I'm pretty sure that the creator of that Checkpoint meant to say "generate" rather than "Train". It's a CivitAI hosted model though. So you could reach out to the creator via comment or chat to get some clarity on what they actually meant by that statement. I'm almost 100% confident they meant to say "generate" though. A lot of times there is a slight language barrier or loss in translation on CivitAI. A lot of Japanese, Filipino, Chinese, and Slovak/Eastern European users/creators there. |
Beta Was this translation helpful? Give feedback.
-
I would like to train on this model: https://civitai.com/models/112902?modelVersionId=354657
It says:
Question is, how do I specify sampler, cfg and target steps to the trainer? I haven't found any of these being configurable somewhere
Thanks!
Beta Was this translation helpful? Give feedback.
All reactions