What Lora Format is this? and how to convert it to diffusers format?

[nighting0le01]

Hi i came accross this lora format for a RV5.1 (SD 1.5 based model) dict_keys(['hookers.layers.0.temporal_i.bias', 'hookers.layers.0.temporal_i.weight', 'hookers.layers.0.temporal_k.bias', 'hookers.layers.0.temporal_k.weight', 'hookers.layers.0.temporal_n.bias', 'hookers.layers.0.temporal_n.weight', 'hookers.layers.0.temporal_o.bias', 'hookers.layers.0.te..
this is from the https://huggingface.co/LayerDiffusion/layerdiffusion-v1/tree/main https://huggingface.co/LayerDiffusion/layerdiffusion-v1/resolve/main/layer_sd15_transparent_attn.safetensors?download=true
how to convert and load this into SD1.5 based models

[sayakpaul]

Are you sure it's LoRA?

[nighting0le01]

@sayakpaul Hi yes it is : it's from LayerDiffusion/layerdiffusion-v1, https://huggingface.co/LayerDiffusion/layerdiffusion-v1/tree/main,
here is the exact LoRA weight: https://huggingface.co/LayerDiffusion/layerdiffusion-v1/resolve/main/layer_sd15_transparent_attn.safetensors?download=true.

orignal repo reference:https://github.com/layerdiffusion/sd-forge-layerdiffuse#:~:text=layer_sd15_transparent_attn.safetensors

[sayakpaul]

Indeed. I think they have additional sources that would be needed to complement such a LoRA checkpoint. More specifically:

• https://github.com/layerdiffusion/sd-forge-layerdiffuse/blob/e4d5060e05c7b4337a3258bb03c4e3ad2f8b15bb/lib_layerdiffusion/models.py#L52
• https://github.com/layerdiffusion/sd-forge-layerdiffuse/blob/main/lib_layerdiffusion/attention_sharing.py

[nighting0le01]

@sayakpaul But those seem to be in LDM format. anyway to apply to a diffusers SD1.5 checkpoint? or get a diffusers implementation

[sayakpaul]

Not yet