It works, but not texture files exportet

[Nde917]

finaly your portable Hunyuan3D-2-WinPortable works (win11, 5090)
exporting meshes works, it exports the mesh but without the texture. There is no error warning.
what could this be?

[SAM-tak]

I’m in the exact same boat. I’ve tried various options provided in the repo, but while I can successfully generate the mesh, I’ve never been able to get the texture to generate.

My environment only has 6GB of VRAM, which is why I’m using this repository. Does this mean that 6GB is only enough for mesh generation and that texture generation is simply not possible?

Has anyone managed to successfully generate and export a model with textures in a 6GB VRAM environment?

[SAM-tak]

An error occurred while trying to fetch C:\AI\Hunyuan3D2_WinPortable\\HuggingFaceHub\models--tencent--Hunyuan3D-2\snapshots\9cd649ba6913f7a852e3286bad86bfa9a2d83dcf\hunyuan3d-paint-v2-0-turbo\vae: Error no file named diffusion_pytorch_model.safetensors found in directory C:\AI\Hunyuan3D2_WinPortable\\HuggingFaceHub\models--tencent--Hunyuan3D-2\snapshots\9cd649ba6913f7a852e3286bad86bfa9a2d83dcf\hunyuan3d-paint-v2-0-turbo\vae.

Yes, it's true that diffusion_pytorch_model.safetensors is missing, and there is only diffusion_pytorch_model.bin instead.
I tried renaming the .bin file to .safetensors, but as expected, that didn't work.

Switching off Turbo Mode didn't make any difference. Even with Turbo Mode disabled, it still tries to reference hunyuan3d-paint-v2-0-turbo\vae\diffusion_pytorch_model.safetensors.

[SAM-tak]

Tencent-Hunyuan/Hunyuan3D-2#345

It seems like this has been an issue since the original version... why hasn't it been fixed yet? Could this be related to why the texture generation isn't working properly?

[SAM-tak]

I downloaded diffusion_pytorch_model.safetensors from https://huggingface.co/tencent/Hunyuan3D-2/tree/main/hunyuan3d-paint-v2-0/vae

Then no errors occur while initializing phase.

But some warnings occur during generation phase:

C:\AI\Hunyuan3D2_WinPortable\python_standalone\Lib\site-packages\diffusers\image_processor.py:148: RuntimeWarning: invalid value encountered in cast
images = (images * 255).round().astype("uint8")
C:\AI\Hunyuan3D2_WinPortable\Hunyuan3D-2\hy3dgen\texgen\utils\dehighlight_utils.py:109: RuntimeWarning: invalid value encountered in cast
image = Image.fromarray((image.cpu().numpy()*255).astype(np.uint8))
C:\Users\Owner\.cache\huggingface\modules\diffusers_modules\local\pipeline.py:598: UserWarning: torch.range is deprecated and will be removed in a future release because its behavior is inconsistent with Python's range builtin. Instead, use torch.arange, which produces values in [start, end).
index = torch.range(29, 0, -bsz, device='cuda').long()
The first timestep on the custom timestep schedule is 989, not `self.config.num_train_timesteps - 1`: 999. You may get unexpected results when using this timestep schedule.
The custom timestep schedule contains the following timesteps which are not on the original training/distillation timestep schedule: [tensor(890), tensor(791), tensor(692), tensor(593), tensor(494), tensor(395), tensor(296), tensor(197), tensor(98)]. You may get unexpected results when using this timestep schedule.

Generated texture is still blank.

[SAM-tak]

huggingface/diffusers#2153

It seems to be an issue specific to the GTX 1660.

I heard that torch.float16 causes problems on this GPU, so I tried changing the code to torch.float32.
However, when I applied it everywhere, the mesh generation crashed. When I limited the changes only to the texture-related parts, the crashes stopped and the warnings disappeared—but the texture remained blank.

I’ve given up on texture generation.