Custom VAE doesn't work

The disk with the previous copy of the program died, so I had to install Python from scratch, then Easy Diffusion itself.

Version 3.0 completely refused to work. Error message:

File "C:\Python\lib\io.py", line 54, in <module>
ImportError: cannot import name 'text_encoding' from 'io' (unknown location)
Failed to run 'conda activate'.

I tried to solve the problem myself, updating the version through the console following advice on the Internet, but even advice from the GPT did not help.

In the end, I reinstalled Easy Diffusion because it listed three options with a link to GitHub, but the advice on that link didn't help either.

Somehow, installing Python on a different drive worked, and I saw the program interface for version 3.0. But it refused to work, so I selected version 3.5 in the settings and restarted the console. After updating the program, it stopped loading after the blue rectangle, displaying a message.

Backend process failed to start (exit code 1)

I followed the link to the interface and updated the program again, this time to version 4.0.

The program launched successfully this time! However, there's some strange behavior compared to the previous version.

1. The console takes a very long time to display performance information, compared to version 3.5...

[sdkit3] [2026-06-09 20:12:04.573] DEBUG   Progress: step 13/21 (61.9%), time: 68.66s
[sdkit3] [2026-06-09 20:13:28.205] DEBUG   Preview callback: step 14, frame_count 1, is_noisy 0
[sdkit3] [2026-06-09 20:13:28.206] DEBUG   JPEG encoding result: 1, jpg_buffer size: 12943
[sdkit3] [2026-06-09 20:13:28.206] DEBUG   Preview: step 14, frames: 1, noisy: 0, jpg_size: 12943 bytes
[sdkit3] [2026-06-09 20:13:28.206] DEBUG   Progress: step 14/21 (66.7%), time: 69.73s

1. When selecting Custom VAE, the output image is white, even though the image is drawn during the generation stage. I tried several different ones, and only the default one works.

2. Some of the "Sampler" and "Scheduler" parameters that were present in the previous version have disappeared from the settings.

3. Text Encoder models are not loading.

I reinstalled the program several times, each time removing the previous version of Easy Diffusion and Python.

Desktop (please complete the following information):

• OS: windows 10
• Browser: Google Chrome 148.0.7778.217 (x64 )
• Version: Easy Diffusion v4.3.0