CUDA out of memory
The model checkpoint needs more VRAM than the card currently has free.
How to fix it
- 01Close other GPU-heavy apps (browser tabs with WebGL, game launchers, other AI tools) and retry the generation.
- 02Switch to a smaller model variant in the model list — the lightweight image-to-3D model fits in about 4 GB of VRAM.
- 03If you are on a low-VRAM build such as “Hunyuan3D 2.1 Full - Low VRAM”, confirm INT8 or FP8 inference is selected.
- 04On laptops with hybrid graphics, set the app to use the discrete GPU if the model keeps failing on the integrated one.