DiffSynth-Studio
DiffSynth-Studio copied to clipboard
modelscope
Enjoy the magic of Diffusion models!
Hi all, I’m encountering OOM issues when using LoRA to finetune Wan2.2-I2V-A14B with the script `lora/Wan2.2-I2V-A14B.sh` on 2×80 GB H20 GPUs. Is this expected? I noticed in the documentation that...
Hi, DiffSynth-Studio team. I am recently playing with wan2.2 i2v training and found that the `token replace` style i2v training is newly adopted in wan2.2. The `input_latents`' is first initialized...
I tried to implement DPO training using LoRA with qwen_image_edit , but it seems that the current code does not support batch inference and also has high GPU memory usage,...
Run Wan2.1-VACE-14B failed when have `vace_video_mask`. error log: ```python3 Downloading Model from https://www.modelscope.cn to directory: DiffSynth-Studio-main/models/Wan-AI/Wan2.1-T2V-1.3B 2025-06-27 09:18:01,016 - modelscope - INFO - Target directory already exists, skipping creation. VAE...
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/examples/wanvideo/model_training/train.py", line 165, in [rank1]: launch_training_task( [rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/trainers/utils.py", line 491, in launch_training_task [rank1]: loss = model(data=None, inputs=data) [rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl [rank1]: return self._call_impl(*args,...
您好，我想请问一下，在训练时high noise和low noise模型使用不同的采样时间边界，为什么推理时switch_DiT_boundary不沿用0.358这个边界而是新值0.875呢？
您好，非常感谢你们的杰出工作 我是一名初学者，基于 Wan2.1-Fun-V1.1-14B-Control-Camera 训练了一个 LoRA ，但lora_base_model是dit，请问 LoRA 是否包含 Camera Adapter 和 Camera Encoder 这两个模块？没包含的话，对lora的相机控制效果是否会有影响？我是否需要将这两个新增模块的参数纳入 LoRA 的优化列表？
example/qwen_image里的lora训练脚本，数据集是data/example_image_dataset，这个数据集只能自己准备吗，还是官方有稳定可以复现的数据集供下载
1. 为什么Wan-animated推理代码和训练代码中的conds的长度都为生成视频长度-4？ https://github.com/modelscope/DiffSynth-Studio/blob/0a1c172a00fb2dd76abedd3b066ddbf62bd4a60d/examples/wanvideo/model_training/validate_lora/Wan2.2-Animate-14B.py#L21 https://github.com/modelscope/DiffSynth-Studio/blob/0a1c172a00fb2dd76abedd3b066ddbf62bd4a60d/diffsynth/pipelines/wan_video_new.py#L1072 2. Wan-animated中训练和推理均没有实现论文中的overlap策略？
