Closed Hazarch closed 7 months ago
The models from the provided SDXL checkpoints additionally learn a different time embedding which is then also being used by the base model.
cheers
The models from the provided SDXL checkpoints additionally learn a different time embedding which is then also being used by the base model.
cheers
Thank you for your prompt reply. The problem has been solved.
When I loaded SDXL and Control checkpoints, the model produced pictures normally. But the model generates noise when I load only SDXL's ckpt. According to the zero layer processing in the paper, should the model be consistent with SDXL at this time?