# SpheRoPE: Amazon Prime Video Researchers Enable Zero-Shot 360° Panorama Generation in Diffusion Transformers

_Research · published 2026-07-06_

Researchers from Amazon Prime Video, Tel Aviv University, and Hebrew University of Jerusalem have introduced SpheRoPE, a zero-shot, training-free framework for generating 360° panoramic images and videos using pre-trained diffusion transformers. Rather than fine-tuning on scarce panoramic datasets or relying on costly per-sample optimization, the method injects spherical geometry directly into existing models at inference time by replacing standard rotary position embeddings (RoPE) with a spherical variant that enforces the topological constraints of equirectangular projection — horizontal periodicity and polar convergence. A companion Semantic Distortion CFG guidance module steers geometry without altering model weights.

The approach is designed to be backbone-agnostic, and the paper demonstrates results across Flux.1, Flux.2, and LTX-Video on both text-to-panorama and image-to-panorama tasks. Because no retraining or architectural changes are required, SpheRoPE inherits the full generative capability of each base model. For VR filmmakers and immersive content creators, the framework lowers the barrier to 360° asset generation considerably, removing the need for specialized panoramic training pipelines.

## Sources
- [orhir.github.io](https://orhir.github.io/SpheRoPE/)
---
Canonical: https://genbuzz.news/posts/spherope-amazon-prime-video-researchers-enable-zero-shot-360-panorama-generation-in-diffusion-transformers