1. Workflow Overview

This workflow generates high-resolution panoramic images using Stable Diffusion XL (SDXL) and UltimateSD Upscale, with features:

• Creates wide-format panoramas (default 2048x1024) with 360° aerial prompts.

• Uses Asymmetric Tiling Sampling to avoid stitching artifacts.

• Enhances details via 4x-UltraSharp upscaling model.

• Applies sharpening before final output.

2. Core Models

3. Key Nodes

4. Workflow Structure

1. Model & Prompt Setup

   • Input: Base model, LoRA, prompts (e.g., "360, panorama, mountains").

   • Output: Encoded conditioning vectors.

2. Panorama Generation

   • EmptyLatentImage sets initial resolution (2048x1024).

   • KSamplerAdvanced generates low-res panorama.

   • Asymmetric_Tiling_KSampler optimizes tiling.

3. Upscaling & Sharpening

   • UltimateSDUpscale upscales tiles (e.g., to 4096x2048).

   • ImageSharpen enhances edge clarity.

   • SaveImage saves the final image.

5. Input & Output

• Input Parameters:

  • Resolution: 2048x1024 (set in EmptyLatentImage).

  • Seed: Fixed or random (Seed node).

  • Prompts: Include "360" or "panorama" (optimized via SDXLPromptStyler).

• Output: Upscaled panorama (PNG), saved twice (initial + sharpened).

6. Notes

• Hardware Requirements: Minimum 12GB VRAM (more needed for upscaling).

• Common Issues:

  • Missing models: Verify paths for wildcardxXL_v4Rundiffusion and LoRA.

  • VRAM errors: Reduce tile_size in UltimateSDUpscale (e.g., to 384).

• Optimization Tips:

  • Launch with --highvram flag.

  • Use "Half Tile" mode in upscaling for speed-quality balance.