1. Workflow Overview

This workflow generates seamless 360° panoramic images for VR, game environments, or architectural visualization. Key features:

• Seamless Stitching: Multi-stage sampling and image blending.

• HDR Support: Enhanced dynamic range for indoor/outdoor scenes.

• AI Upscaling: Uses 4x-UltraSharp to achieve 4K output.

Core Models:

• Flux Fill dev - fp8: Optimizes panorama details.

• 360全景Lora_360HDR: Custom LoRA for HDR and perspective consistency.

• UMT5-XXL: Multilingual text encoder (supports Chinese prompts).

2. Key Nodes

Dependencies:

• Manual downloads:

  • flux1-schnell-Q8_0.gguf (UNet model).

  • 360全景Lora_360HDR.safetensors (LoRA weights).

3. Workflow Groups

1. Loader: Loads models, VAE, and text encoders.

2. Settings: Configures global seed and latent size.

3. Conditioning: Processes prompts with FluxGuidance.

4. Noise Sampler: Generates latent images in regions.

5. Crop Image: Splits images into 256x768 tiles.

6. Inpaint: Fixes seams using Florence2Run.

7. Build Image: Stitches tiles and upscales.

8. Save Image: Exports 4K panorama.

4. Inputs & Outputs

• Inputs:

  • Prompts: e.g., "futuristic bedroom in a cave, 360HDR".

  • Resolution: Default 1536x768 (2:1 aspect ratio).

• Output:

  • 4K 360° panorama (PNG).

5. Notes

• VRAM: ≥16GB GPU recommended (use fp8 for lower usage).

• Debugging:

  • Missing models → Check upscale_models and loras paths.

• Optimization:

  • Generate at lower resolution (e.g., 1024x512) then upscale.

  • Set FluxGuidance ≤5 to avoid over-sharpening.