ComfyUI-Furrey-TurnaroundSplitter

🎯 Furrey-TurnaroundSplitter for ComfyUI

Automatically splits a character turnaround sheet (front/side/back/side, or any number of side-by-side views) into separate cropped images — no manual pixel math required.

Built for multi-view image-to-3D pipelines (Trellis2 / Pixal3D, Hunyuan3D, TripoSR, etc.) that need each view of a turnaround sheet fed in separately, but works with any sheet of flat-background images placed side by side.

Also includes Scale Mesh (per axis), a small utility to fix a squashed/stretched 3D output by imposing the proportion you want directly in the workflow — see below.

✨ Features

• 🔍 Auto-detects the background color — works with black, white, or any flat/uniform background, no need to specify it.
• 📐 Auto-detects each view's boundaries — finds the gaps between figures and crops around them, regardless of image size, aspect ratio, or how many views are in the sheet.
• 🧩 Handles uneven views — each cropped output keeps its own width (a front/back pose is usually wider than a side profile), matching what multi-view 3D pipelines expect.
• 🛠️ Tunable — padding, background sensitivity, minimum gap/width and an optional tight vertical crop are all exposed as widgets for tricky sheets.
• 📝 Debug output — an info text output reports exactly which box was used for each view, so you can see what was detected without guessing.

📦 Installation

Via ComfyUI Manager (Recommended)

1. Search for "Furrey-TurnaroundSplitter" (once indexed).
2. Click Install.

Manual Installation

1. Go to your ComfyUI/custom_nodes/ directory.
2. Clone this repo:

   git clone https://github.com/SilentLuxRay/ComfyUI-Furrey-TurnaroundSplitter
   

3. Restart ComfyUI.

🧠 How it works

The node samples the image's border pixels to learn the background color, then measures how much each column of the image differs from that color. Columns that are mostly background mark the gaps between figures; contiguous "content" columns become one detected view. Each view is then cropped out (with your chosen padding) at its own natural width — nothing is stretched or resized to force a uniform size.

⚙️ Node reference

Auto-Split Turnaround Sheet (category image/transform)

| Input | Description | |---|---| | image | The turnaround sheet to split | | num_views | How many views to detect/output (1-8) | | padding | Fixed extra margin (px) added around each detected view. Ignored if margin_percent > 0 | | margin_percent | Margin as a % of the view's own size, added on every side (self-scales to any resolution). Many 3D/conditioning pipelines expect the subject to occupy ~90% of the frame (~8-10% margin), not edge-to-edge — 0 disables this and falls back to padding | | bg_threshold | Per-pixel distance from the detected background color to count as "content" | | min_col_fraction | Minimum fraction of a column that must be "content" to count it as part of a figure (filters noise) | | merge_gap | Background gaps narrower than this (px) get merged into the same view, instead of splitting it | | min_width | Minimum width (px) for a detected block to count as a real view, not noise | | tight_vertical | Crop each view to its own content height instead of the full image height | | pad_to_square | Pad each cropped view to a square canvas, centered, filled with the detected background color — useful for pipelines (e.g. CLIP-Vision-based conditioning) that expect square, centered inputs |

Outputs: view_1 … view_8 (unused slots return an 8×8 black placeholder) and info (a text summary of what was detected, for debugging).

🖼️ Example

Feed a LoadImage of a 4-view turnaround sheet straight into this node, num_views=4, and wire view_1..view_4 into whatever your pipeline expects per view (RemoveBackground, crop-to-mask, conditioning, etc.) — no manual x/y/width/height guessing.

📐 Bonus node: Scale Mesh (per axis)

A small utility node (category 3d/mesh) that scales a MESH's vertices independently per axis (X/Y/Z) around the bounding-box center (or world origin). Useful when a 3D generator's output comes out squashed or stretched on one axis and there's no exposed parameter to fix it upstream — impose the correct proportion directly in the workflow, right after mesh generation and before saving, instead of fixing it by hand in Blender every time.

| Input | Description | |---|---| | mesh | The mesh to scale | | scale_x / scale_y / scale_z | Per-axis multiplier (1.0 = unchanged) | | pivot | bbox_center (scale around the mesh's own bounding-box center) or origin (scale around world 0,0,0) |

---

Created by Furrey for the ComfyUI community.