-
|
I tested and trained it repeatedly in onetrainer and it was completely normal in sampling, but once I used comfyui the output was completely ruined, the same model did not work properly. The lora trained on Civitai is working perfectly fine. I don't understand where I set it wrong? Is it related to bf16 or fp16? I can't find any effective answer, on the contrary, there are a lot of useless ads... And I have asked the AI many times and experimented for many days, but I can't solve the problem. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
You have neglected to mention the model..so I will assume Huyuan, this is not an OT issue. The issue lies with Comfy. Revist this in a month and it hopefully will be solved. |
Beta Was this translation helpful? Give feedback.
-
|
By switching to Akegarasu's lora-scripts my SDXL lora training became normal and of good quality. Onetrainer sampling process is good, but the actual output is extremely bad. I wasted a week and couldn't get it to work properly. I no longer use it. |
Beta Was this translation helpful? Give feedback.
You have neglected to mention the model..so I will assume Huyuan, this is not an OT issue. The issue lies with Comfy. Revist this in a month and it hopefully will be solved.