GPT Image Bridge · Edit Reference List (Image 2–9)

GPT Image Bridge · Edit Reference List (Image 2–9) is the pack's most boring node, and it's fine that it is. It exists to solve one problem: getting multiple reference images into the Edit node in a stable, ordered way.

What it does

The naming is doing a lot of work, so read it carefully. The Edit node's roles are fixed - Image 1 is the base image being edited, and this Reference List node supplies Images 2 through 9. So its eight inputs are image_2, image_3, all the way up to image_9, each accepting a single IMAGE. It packs them into one references output (type GPT_IMAGE_REFERENCES) that plugs into the Edit node's references（Image 2 开始） input.

Why it exists

Why a dedicated node instead of just feeding images straight in? Order and encoding. Each reference is encoded independently, which means your references can be different sizes - no pre-resizing, no forced IMAGE batch. And the ordering is structural: image_2 is the first reference, image_3 the second, and so on, so canvas order, report order, and the order the remote request sends them all stay in lockstep. The whole point is that GPT Image is sensitive to which image is which, and this node makes it impossible to scramble that by accident.

Two rules enforce the ordering, and they're the ones beginners hit:

• No gaps. If you connect image_2 and image_4 but skip image_3, the node refuses with a "connect without gaps" error before anything hits the network.
• One image per slot. A batch fed into a single input is rejected - the error tells you to spread the batch across the numbered slots.

There are no other inputs or settings, and no outputs beyond references. Single-image edits don't need this node at all - just connect base_image（Image 1） to the Edit node and leave the references input empty. And a point the README is explicit about: this node is for Edit only. Generate has its own direct reference_1–reference_9 inputs, so don't reach for the Reference List there.

Install

Install is the shared pack story: ComfyUI Manager (search GPT Image Bridge) or git clone https://github.com/Liu-Bot24/comfyui-gpt-image-bridge.git ComfyUI-GPT-Image-Bridge into custom_nodes, pip install -r requirements.txt (just numpy and Pillow), restart. No models, no keys - the node holds no configuration, which is exactly why it can't break in interesting ways. Its only failure modes are the gap and batch checks, and those fail loudly and locally before you spend any API credits. It's the unglamorous plumbing the rest of the pack depends on.