kiui

@deividmor Hi, you may try to disable fp16 inference by changing `fp16` to `False` here: https://github.com/dreamgaussian/dreamgaussian/blob/main/guidance/zero123_utils.py#L22C34-L22C34

Windows environment can be tricky, these errors may indicate that MSVC and CUDA versions are not matched.

Hi, could you check `torch.cuda.is_available()`?

@emailandxu Hi, that's typical for the multi-view images generated by ImageDream, maybe due to the bias in their training dataset. You may try other random seeds, which may not have...

@Ysz2022 Hi, could you try to re-install the `diff-gaussian-rasterization` package and check the logs?

Yes, actually the model is trained with 4 random views. You could check this https://github.com/3DTopia/LGM/blob/main/core/models.py#L61 to specify the camera poses.

@cwchenwang You need to make sure the pose is in OpenGL world system & camera convention (https://kit.kiui.moe/camera/). I remember the poses from their json are in blender world system.

 Sure, it's actually generated by Fooocus.

@rajatprasun Hi, unfortunately MVDream can be unstable sometimes. You can try with different random seeds, and examine the mv images first. The mesh is exported in GLB, which is supposed...

@rajatprasun Surely the model still have some limitations for complex prompts... This ply file is the 3D Gaussians instead of the final mesh. You need to run `convert.py` on this...