comfyui-egregora-regional 🧩🌾

✨ Image Tile Split with Region-Aware Prompting for ComfyUI

---

🚀 Features

🧱 Egregora Tile Splitter

• Flexible grid with overlap (aligned with Tiled Diffusion).
• Emits a shared grid_json used by the other nodes so everything stays in sync.
• Edge-safe alignment for SDXL/SD1.5 latent grids.

🗺️ Egregora Regional Conditioning (Unified)

• Per‑tile positive/negative prompts from lists or text.
• ✅ Accepts Primitive → String (multiline) directly (e.g., output pasted from Florence‑2).
• Blends tile‑specific text with optional global base prompts.
• Smooth feathered masks to reduce seams across tiles.
• Validates and snaps each region to the latent grid to avoid artifacts.

🤝 ControlNet Compatibility

• Combine with Apply ControlNet or Apply Advanced ControlNet.
• Use Conditioning (Average) to mix global ControlNet guidance with regional prompts.
• Prevents blurry results by blending instead of stacking conditionings.

---

📦 Installation

1) Via ComfyUI Manager (recommended)

1. Open ComfyUI → Manager.
2. Select Install from URL and paste this repo URL:

https://github.com/lucasgattas/comfyui-egregora-regional

3. Click Install and then Reload ComfyUI.

2) Manual Installation

cd ComfyUI/custom_nodes
git clone https://github.com/lucasgattas/comfyui-egregora-regional.git
# restart ComfyUI

---

🧩 Nodes

🖼️ Egregora Tile Splitter

Splits an IMAGE (B,H,W,C) into overlapping tiles and returns:

• tiles_batch – concatenated tiles
• grid_json – grid metadata for downstream nodes

---

✍️ Egregora Regional Conditioning (Unified)

Builds per‑tile CONDITIONING (positive & negative) from lists or text. Supports:

• tile_prompts_json – one prompt per tile (list or newline text)
• base_positive, base_negative with adjustable strengths
• CLIP text‑encode cache for speed
• Feathered tile masks to minimize seams
• Primitive → String (multiline) input (ideal for Florence‑2 output)

Inputs: clip, grid_json, tile_prompts_json Outputs: conditioning_pos, conditioning_neg

---

⚡ Quick Start

1. Resize your input image to target resolution.
2. Egregora Tile Splitter → set tile_w, tile_h, overlap.
3. Florence‑2 (or other captioner) → produce per‑tile text → Primitive:String(multiline) → wire into Egregora Regional Conditioning.
4. Egregora Regional Conditioning → connect the same grid_json and clip.
5. CLIP Text Encode (base prompt, optional).
6. Apply ControlNet (Tile, Canny, Depth, etc.) on the base conditioning.
7. Conditioning (Average) → blend Regional Conditioning with ControlNet‑augmented conditioning.
8. Tiled Diffusion → set the same tile_width, tile_height, tile_overlap.
9. Sampler → LCM or other sampler of your choice.
10. VAE Decode (Tiled) → decode as usual.

Upscaler example (high‑level):

IMAGE ➜ Tile Splitter ➜ Florence‑2 ➜ Primitive:String(multiline) ➜
Regional Conditioning ➜ CLIP Text Encode ➜ Apply ControlNet ➜
Conditioning (Average) ➜ Tiled Diffusion ➜ KSampler ➜ VAE Decode

---

💡 Tips

• Bigger tiles + smaller overlap = faster; more overlap = smoother transitions.
• Match Tile Splitter and Tiled Diffusion overlap for seam‑free blending.
• Start with Conditioning (Average) strength ~0.3–0.5 for regional prompts.
• Keep ControlNet strength moderate (0.3–0.6) to avoid over‑constraint.

---

✅ Compatibility

• ComfyUI (latest stable)
• Works with SDXL and SD1.5
• No additional Python packages required

---

📝 License

MIT

---

🙌 Credits

Inspired by the ComfyUI community, Tiled Diffusion, and ControlNet workflows. Special thanks to those pushing regional prompting and advanced upscalers forward.