Native ComfyUI nodes for seamless 360-degree equirectangular panorama generation with native FLUX.1 and Wan video models, adapted from SpheRoPE.
The native path is designed to fit an existing ComfyUI graph: it patches a regular FLUX.1 or Wan MODEL, packages ERP conditioning into ordinary positive/negative CONDITIONING, works with the standard KSampler, and returns a wrapped VAE that can be connected to the standard VAE Decode node.
Non-commercial use only. This repository follows the upstream CC BY-NC 4.0 license. See LICENSE and THIRD_PARTY_NOTICES.md.
Clone this repository into ComfyUI/custom_nodes and restart ComfyUI:
cd ComfyUI/custom_nodes
git clone https://github.com/jay0n9/ComfyUI-SpheRoPE.gitNo additional Python packages are required for the native nodes beyond
ComfyUI's existing torch, numpy, and Pillow stack. The included
requirements.txt is intentionally package-free so ComfyUI Manager does not
replace the environment's CUDA-enabled PyTorch build.
- Load a FLUX.1 or native Wan model, CLIP, and VAE as usual.
- Connect
MODELandVAEto SpheRoPE Pipeline Patch. - Connect the patched
spherope_modelto a standard KSampler. - Use SpheRoPE ERP Conditioning instead of separate positive/negative CLIP text encode nodes.
- Connect its positive and negative outputs to the standard KSampler.
- Decode the sampled latent with a standard VAE Decode, using
circular_vaefrom SpheRoPE Pipeline Patch. - Generate at a 2:1 equirectangular resolution such as 1024×512 or 2048×1024.
The ERP geometry prompt is intentionally internal and fixed:
Single unified continuous environment, monolithic scene composition, solitary spatial layout, flawlessly stitched 360 panorama, true equirectangular projection, accurate spherical geometry, continuous horizontal wrap, zero parallax error.
erp_gamma controls the added ERP branch. A practical starting value is 4.0; set erp_enabled to false to disable that branch without changing graph wiring. Regular KSampler CFG remains available as usual.
An importable example is included at examples/spherope_native_flux1_modular.json.
| Node | Purpose |
|---|---|
| SpheRoPE Pipeline Patch | Combines the native Wan/FLUX.1 spherical-frequency-coordinate model patch and circular VAE wrapper. |
| SpheRoPE ERP Conditioning | Encodes positive/negative prompts and embeds the fixed ERP branch plus erp_gamma metadata. |
Only these two integrated nodes are registered in ComfyUI. Legacy and internal helper implementations remain unregistered to keep the node menu and workflow surface minimal.
The native SFC model patch supports FLUX.1 (axes_dim == [16, 56, 56]) and native ComfyUI Wan models with three-axis RoPE. For Wan VACE, connect ModelSampling -> SpheRoPE Pipeline Patch -> Mobius Model Patch -> KSampler.
This project is an independent ComfyUI integration/port based on the ideas and implementation of:
- orhir/SpheRoPE
- SpheRoPE: Zero-Shot Optimization-Free 360 Panorama Generation with Spherical RoPE, Hirschorn et al., arXiv:2606.32033
Changes in this repository include native ComfyUI model hooks, standard-KSampler ERP conditioning, a combined model/VAE patch node, a circular VAE proxy, and ComfyUI workflow integration. This repository is not affiliated with or endorsed by the upstream authors.
The code is distributed under Creative Commons Attribution-NonCommercial 4.0 International to preserve the upstream terms. Model weights and other third-party components may have separate licenses; you are responsible for complying with them.
Use sfc_strength = 0.25 and erp_gamma = 0.05 with CausVid CFG 1.0. Gamma values around 0.5 or above can dominate four-step Wan denoising and collapse the image into horizontal bands. Increase gamma only in small increments after an A/B test.