ChaiNNer Workspace

---

This repository contains a custom workspace file (.chn) for chaiNNer, focused on high-quality upscaling with intelligent segmentation.

The key feature of this workflow is its ability to separate the entire person (body and face) from the background, treating each layer with specific AI models to achieve the best possible result, reconstructing the final image without losing naturalness.

✨ Key Features

• Hybrid (Video and Image): A single file serves both purposes with a simple selector.
• Full Body Segmentation: Unlike workflows that focus only on faces, this uses advanced models to cutout the complete human silhouette.
• Dedicated Processing:
  • Person/Face: Processed with models focused on skin and facial features.
  • Background: Processed with models focused on realistic scenarios and textures.
• Smart Resizing: Option to define a Target Height, automatically maintaining the aspect ratio.

---

⚠️ Important Requirement (Avoiding Errors)

Due to how the chaiNNer validator works, you must load files into BOTH input nodes (Image and Video), even if you are only going to process one type.

If you leave one of the nodes empty, chaiNNer will prevent execution with the following errors:

• Image: Load Image: Missing required input data: Image File • Image: Load Video: Missing required input data: Video File

Solution:

• If processing a Video: Load your video into the video node and any "dummy" image (any jpg/png) into the image node.
• If processing an Image: Load your image into the image node and any "dummy" video (any short mp4) into the video node.

---

⚙️ How to Use (Mode Selector)

The workspace has been updated with logic to easily switch between Image and Video modes without manually disabling nodes.

1. Setup Inputs

Load your files into the Load Image and Load Video nodes (remember the warning above: both nodes must have files loaded).

2. Select the Type (Node "SELECT THE TYPE")

Locate the group of nodes with the yellow note "Select Input". In the "Select the Type" node:

• Change the "Value Index" to:
  • A 📸 → To process IMAGE.
  • B 🎥 → To process VIDEO.

3. (Optional) Define Size

In the "Target size" group, change the value in the "Target Height" node:

• Set the desired height in pixels (e.g., 1080 for FHD).
• Leave at 0 to disable resizing and keep the original size (or the native upscale size).

---

🧠 Models Used

To ensure the best quality, make sure you have the following models:

1. Person/Face Upscale

Model: 4xFaceUpDAT
Successor to 4xFFHQDAT, trained on the FaceUp dataset.

• Recommendation: Use the 4xFaceUpDAT version.

2. Background Removal (Segmentation)

Model: briaai/RMBG-1.4
State-of-the-art background removal model to precisely isolate the full human figure.

3. Background Upscale

Model: 4x-ClearRealityV1
Trained on the SPAN architecture, this model focuses on realistic imagery and natural scenery.