Compatibility issues regarding checkpoint/VAE dependency conflicts when Diffusers load Civitai LoRA

[hhhFuture]

Hello everyone, I'm currently learning to use diffusers and would like to ask all my friends a question. I saw a good lora on Civitai, but this lora has requirements for checkpoint and vea. So I downloaded both models as the author requested. However, when I ran the following code, an error occurred.
The specific code is as follows:

pipeline = StableDiffusionPipeline.from_single_file(
    r"E:\Project_draw\Models\vae\clearvaeSD15_v23.safetensors",
    use_safetensors=True,
    torch_dtype=torch.float16,
    safety_checker=None
 )

The errors are as follows:

I checked the documentation of diffusers. The documentation mentioned that it is possible to load the model in this way, but I don't know why an error occurred. I saw many convert scripts in the script folder of diffusers, but I don't know which one is the corresponding conversion script and what the requirements are. If there are any friends who know how to solve it, could you please tell me what it feels like

[asomoza]

Hi, this is a question and not an issue, so I'm moving it to the appropriate section.

About your question, you don't provide the models so it's hard for us to test it, but it seems you're loading a vae, if it's in the original format, you should load it like this:

from diffusers import AutoencoderKL

vae = AutoencoderKL.from_single_file(url, ...)

[hhhFuture]

Thank you very much for your reply. Due to my personal negligence, I forgot to post the link. Please excuse me, my friend. Here is the link.
lora:https://civitai.com/models/54233/ghiblibackground?modelVersionId=125985

Lora corresponding eva:https://civitai.com/models/22354

Lora corresponding checkpoint:https://civitai.com/models/4468?modelVersionId=7425

Following the author's advice, I downloaded these three models to my local device and used them according to the interfaces in the diffusers tutorial. Except for lora, all the others reported errors. I looked up relevant materials, which mentioned: The models on Civitai cannot be directly invoked by diffusers because all the models on Civitai are trained for webui and are not applicable to diffusers. Conversion is required before they can be used. I'm not sure if it's caused by this reason. If conversion is needed, where can I find relevant tutorials or documents on conversion? And I saw that many lora information introductions on Civitai mentioned the base model :SD1.5. So, can I understand that the base model of lora is based on the SD1.5 system? The specific base model still needs to be downloaded and cannot be directly used with SD1.5, right?I'm looking forward to your reply very much. Thank you again.

I tried numerous methods today, but all failed. I checked related issues and found similar problemshttps://github.com//issues/4436, but they still haven't been solved. Among the methods I tried today are:

1. When using the convert_vae_pt_to_diffusers.py provided by the diffusers official for conversion, although an additional config was added after the conversion, attempting to call the converted safetensors through the interface still failed. The specific terminal results are as follows:
   cmd screenshot

The safetensors and config generated after conversion:

Screenshot of an error when using the converted model:

2. Using AutoencoderKL.from_single_file() also failed. The specific operation is as follows:

I was extremely eager to solve these problems and was confused about them for several days. I hope to receive help from all my friends. Of course, it would be best if there were suitable documents to recommend. I also sincerely ask my friends to help solve the problems I am currently facing. Thank you all in advance.