there’s a new model based on illustrious trained on v-pred ztsnr:https://civitai.com/models/833294/noobai-xl-nai-xl, would be nice if the site trainer gave support for image generation, and lora training, i generally use the args --v_parameterization --zero_terminal_snr --scale_v_pred_loss_like_noise_pred --debiased_estimation_loss and disable any noise related things like min snr gamma to train a lora, i dont think it should be that hard to give support.
Please authenticate to join the conversation.
Awaiting Dev Review
💡 Feature Request
Over 1 year ago

Darkwen
Get notified by email when there are changes.
Awaiting Dev Review
💡 Feature Request
Over 1 year ago

Darkwen
Get notified by email when there are changes.