metadata
license: openrail
Can be used in sd-webui-controlnet.
Put the yaml file with model because it is sd2.1 v_prediction model.
Model is the difference from the original checkpoint created by extract_controlnet_diff.py.
diff_control_wd15beta2_canny.safetensors
It seems pretty good.
- target model: https://huggingface.co/waifu-diffusion/wd-1-5-beta2
- dataset: 11k 1girl images
- epoch: 2
- batch size: 16
- lr: 5e-5
- amp:bfloat16
diff_control_wd15beta2_pose.safetensors
It seems a little subtle.
- target model: https://huggingface.co/waifu-diffusion/wd-1-5-beta2
- dataset: 14k umamusume images. Therefore, the model may have a bias
- epoch: 5 (The “sudden convergence phenomenon” occurred at the fourth epoch)
- batch size: 16
- lr: 5e-5
- amp:bfloat16