ModuleNotFoundError when importing diffsynth.pipelines.qwen_image - Circular Import Issue

I'm trying to train a Qwen-Image LoRA using the provided training scripts, but encountering a module import error
when running the training command.

Environment:

• OS: Windows 11
• Python: 3.11
• DiffSynth-Studio: Latest from main branch
• Installation method: pip install -e .

Steps to Reproduce:

1. Clone the repository: git clone https://github.com/modelscope/DiffSynth-Studio.git
2. Install the package: cd DiffSynth-Studio && pip install -e .
3. Run the training script: accelerate launch examples/qwen_image/model_training/train.py

Error Traceback:
Traceback (most recent call last):
File "E:\dif\DiffSynth-Studio\examples\qwen_image\model_training\train.py", line 2, in
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
File "E:\dif\DiffSynth-Studio\diffsynth_init_.py", line 2, in
from .models import *
File "E:\dif\DiffSynth-Studio\diffsynth\models_init_.py", line 1, in
from .model_manager import *
File "E:\dif\DiffSynth-Studio\diffsynth\models\model_manager.py", line 4, in
from .downloader import download_models, download_customized_models, Preset_model_id, Preset_model_website
File "E:\dif\DiffSynth-Studio\diffsynth\models\downloader.py", line 6, in
from ..configs.model_config import preset_models_on_huggingface, preset_models_on_modelscope, Preset_model_id
File "E:\dif\DiffSynth-Studio\diffsynth\configs\model_config.py", line 76, in
from ..models.qwen_image_text_encoder import QwenImageTextEncoder
[Error continues with circular import chain]

Analysis:
This appears to be a circular import issue where:

1. diffsynth.init.py imports from models
2. models.model_manager imports from downloader
3. downloader imports from configs.model_config
4. model_config tries to import back to models.qwen_image_text_encoder

This creates a circular dependency that prevents the module from loading properly.

Questions:

1. Is the current training setup supposed to work out of the box?
2. Are there any missing dependencies or installation steps?
3. Could this be related to the recent Qwen-Image FP8 updates mentioned in recent commits?
4. Should the imports in model_config.py be moved or made conditional to avoid circular dependencies?

Attempted Solutions:

• Reinstalled with pip install -e .
• Tried different Python versions
• Ensured all dependencies are installed
• Verified the repository is up to date

Any guidance on resolving this import issue would be greatly appreciated. Thank you!

Additional Context:
I'm specifically trying to train a custom LoRA on a personal dataset following the examples provided in
examples/qwen_image/model_training/. The goal is to fine-tune Qwen-Image for specific image generation tasks.

Start by running accelerate launch examples/qwen_image/model_training/train.py and follow the import chain through diffsynth/__init__.py, diffsynth/models/model_manager.py, diffsynth/models/downloader.py, and diffsynth/configs/model_config.py. Check how model_config.py imports QwenImageTextEncoder and identify the dependency cycle. Done means the Qwen-Image training entry point imports without the reported circular-import error.