[bug]: Klein Model's not properly identifying vae/Qwen source

[m4iccc]

Is there an existing issue for this problem?

• I have searched the existing issues

Install method

Invoke's Launcher

Operating system

Windows

GPU vendor

Nvidia (CUDA)

GPU model

3090

GPU VRAM

24

Version number

6.12.0

Browser

No response

System Information

Whenever I download a klein model from your default model downloads, it downloads in diffusers, I assume it also downloads the qwen text encoder and vae, however it doesn't reflect on the frontend, and it loads correctly, however the issue comes when I download a custom Klein model and try to use a vae or qwen text encoder with it, 1. If I try installing the flux 2 klein vae or text encoder the klein 9b diffuser model fails to generate-load now, because it will try to use the newly installed vae and qwen. 2. The new model also fails to load the vae and text encoder, I made sure to download the appropiate variants of qwen and the vae for klein, I'm not sure if a clean install fixes this issue, however invoke should support a way to migrate your models between versions so that these features don't break and you can keep your database intact. I did try a clean install on a separate invoke instance and it doesn't seem to fix the issue, as you can see, when you download the 9B diffuser variant from your default downloads the vae and text encoder are empty, and even though you download the correct variants and install them separately in the model manager, if you try to generate then it will prompt an error and not generate.

What happened

as mentioned above

What you expected to happen

Please fix this issue!

How to reproduce the problem

No response

Additional context

No response

Discord username

No response

[Pfannkuchensack]

So the Diffusers version has both the VAE and text encoder included because that is how the format works.
The z-image Model type has the thing that you could use a diffusers vae or/and text encoder with another base model, for the Flux2klein UI we did not do that because it is confusing.
You need to download a seperated VAE and text encoder, if you use a non diffusers model.