Reference Chain Conditioning

ComfyUI-ReferenceChain

A ComfyUI custom node that simplifies working with edit models (like Flux-2-klein) that accept multiple reference images. Instead of chaining multiple "Reference Latent" nodes together, this node lets you add any number of images through a single, easy-to-use interface.

Features

• Multiple image input - Add as many reference images as you need through a visual list interface
• Drag and drop - Drop images directly onto the node
• Built-in scaling - Automatically scales images to target megapixels before encoding
• API-friendly - Includes a Base64 variant for programmatic usage
• Native styling - Matches ComfyUI's look and feel

Nodes

Reference Chain Conditioning

The main node for interactive use. Add images through the UI, and they'll be automatically processed and chained into a single conditioning output.

Inputs:

• conditioning - Input conditioning to append reference latents to
• vae - VAE model for encoding images to latents
• upscale_method - Scaling algorithm (nearest-exact, bilinear, area, bicubic, lanczos)
• scale_megapixels - Target size for images before encoding (default: 1.0 MP)

Outputs:

• conditioning - Modified conditioning with all reference latents
• first_image_scaled - The first image after scaling (useful for preview)

Reference Chain Conditioning (Base64)

Same functionality but accepts base64-encoded images. Ideal for API workflows.

Input format:

["data:image/png;base64,iVBORw0KGgo...", "iVBORw0KGgo..."]

Or a single base64 string without JSON array wrapper.

Installation

Via ComfyUI Manager

Search for "ReferenceChain" in the ComfyUI Manager and click Install.

Manual Installation

cd ComfyUI/custom_nodes
git clone https://github.com/remingtonspaz/ComfyUI-ReferenceChain.git

Restart ComfyUI after installation.

Usage

1. Add the Reference Chain Conditioning node to your workflow
2. Connect your conditioning and VAE inputs
3. Click "choose file to upload" or drag images onto the node
4. Connect the output conditioning to your sampler

The node processes images in order, scaling each to the target megapixels, encoding through the VAE, and appending to the conditioning - equivalent to manually chaining multiple Reference Latent nodes.

Requirements

No additional dependencies - uses only libraries already included with ComfyUI (numpy, torch, Pillow).

License

MIT