Empty LoRA / Zero-Weight LoRA
This is a valid LoRA file where all weights are set to 0.0.
It is not meant to change the image. It is mainly useful for testing workflows, XY grids, LoRA comparison setups, scripts, or any situation where you need a real LoRA file loaded, but want it to have no visual effect.
This is not the same as making an empty .txt file and renaming it to .safetensors.
A renamed text file is not a valid LoRA and will not load correctly in most tools.
Use this only when you specifically need a neutral/blank LoRA placeholder.
---
空 LoRA / 零权重 LoRA
这是一个有效的 LoRA 文件，里面的所有权重都设置为 0.0。
它不会改变生成出来的图片。它主要用于测试工作流、XY 对比图、LoRA 对比测试、脚本测试，或者任何需要加载一个真实 LoRA 文件、但又不希望它影响画面的情况。
这**不是**把一个空的 .txt 文件重命名为 .safetensors。
重命名的文本文件不是有效的 LoRA，大多数工具都无法正确加载。
只有在你确实需要一个中性/空白 LoRA 占位文件时才需要使用它。
Description
Comments (10)
what 'empty lora' for !? , if used it (even if there any use case lol)let it at 0 ?
It is clearly stated in the description of the LoRA.
It can be usefull when making a grid in Comfyui to mention one example i have used it for multiple times.
And as i also stated, the weights is 0.0, so it is not trained at all. It is just a completely empty LoRA that wont affect the image at all. No usecase for a normal user, but can be very usefull for a select few people.
@Loraholic thanks for your reply, i have read the description and your reply and i don't understand the use case of this . very useful for select few people !? what grid meant here.. still can't understand what is this for.
@amazingbeauty For example when i made this image, i used the empty lora on row 1. That was the only way i could make that image like i wanted with the setup i had. So this is a ACTUAL image that i have uploaded to civitai before i published this LoRA for others.
https://civitai.red/images/130397274
just create an empty txt file and rename it .safetensors?
I dont think that would work. The point is to have a valid file to load. A empty txt is not a valid safetensors file.
At minimum some metadata is needed by safetensors, and one up / down layer to be identified as a lora. Maybe even several or specific ones, as shapes / names might be used for base identification.
What might be minimal in terms of storage is a rank 0 lora: np & tensor both technically support a shape of [d, 0, u] / [u, 0, d], which in an additive capacity should yield the identity function, but it'd be interesting to see whether it works as intended.
I have not tried a to make a Rank 0 tbh X)
This LoRA was a Rank1 LoRA that had training data on it, but changed all lora_up and lora_down to 0,0 in a tool i have to adjust layers manually.
So there is probably some metadata left in it thats saying it has some training steps etc, but the actual numbers are all 0,0.
@Loraholic ah gotcha. i use my txt trick when i need to fill places somewere, yes obviously you get a lora loading error message but the workflow wont stop