MoVerse Pano DiT — AoTI kernels
AOTInductor-compiled repeated blocks for the LoRA-fused FLUX.1-Fill-dev transformer of
hugging-apps/moverse (MoVerse Stage I,
Orange-3DV-Team/MoVerse gimbal360 LoRA fused at scale 1.0).
Built by
hugging-apps/moverse-aoti-compile.
import spaces
spaces.aoti_blocks_load(pipe.transformer, "hugging-apps/moverse-pano-aoti")
The archives contain compiled kernels only — no weights. Weights are bound at load time from
the live module's state_dict(), so the serving Space keeps its own checkpoint and the eager
path stays intact.
| Hardware | ZeroGPU · NVIDIA RTX PRO 6000 Blackwell (sm_120) |
| torch | 2.11.0 |
| diffusers | 0.39.0 |
| dtype | bfloat16 |
| Dynamic | image sequence length (panorama height 512–1024, 2:1 ERP) |
| Text sequence | 512 (static) |
Kernels are hardware- and version-specific: a Space loading them must run the same ZeroGPU hardware and the same pinned torch / diffusers.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for hugging-apps/moverse-pano-aoti
Base model
black-forest-labs/FLUX.1-Fill-dev