Burve Image Reference Pack

A tiny utility that does exactly one job: take up to fourteen images and bundle them into the single IMAGE_LIST that Burve Google Image Gen and the Vertex twin expect on their reference_images socket.

Why does that need a node at all? ComfyUI's standard image types are single tensors or batches, and the Gemini nodes want a list of images - each one potentially a different size, with its own frames. That's not a native shape in most of the graph, so the pack needs a place to assemble it. This is that place.

It's the plumbing for one of the genuinely good things about Gemini image models: they take reference images natively, up to 14 of them. Character consistency is the single most-discussed unsolved problem in local diffusion, and it's usually attacked with adapters and LoRAs. Here, the model just looks at your reference while it draws - which is a different and often more reliable mechanism, and it's what the pack's whole character workflow leans on.

The inputs and outputs

Fourteen optional IMAGE inputs, unimaginatively named image1 through image14. Hook Load Image nodes (or anything else that emits an image) into the slots you need. The node keeps each input's batch as-is, skips the empty slots, and returns one images output of type IMAGE_LIST.

From there it wires straight into the generator's reference_images input. One gotcha: since every slot is optional, an all-empty run gives you an empty list - connect at least one image.

Using it in a workflow

The simple version is a ref-pack for style: load a few outfit or moodboard images, pack them, feed the generator. The more interesting version is the pack's character workflow, where Burve Character Planner does its own reference packing internally - you'd use this node for the extra non-face references, routed into the planner's extra_reference_images input rather than directly at the generator.

Install and gotchas

Nothing special here - install the pack (ComfyUI Manager, search "ComfyUI_Burve_Tools", or git clone https://github.com/Burve/ComfyUI_Burve_Tools.git into custom_nodes), and the node appears under BurveTools. It's pure tensor shuffling, so it doesn't add any heavy dependencies on its own; the pack's google-genai and pillow requirements come along regardless. One thing to know: an IMAGE_LIST from this node only plugs into sockets typed for that custom type, so it's a Burve-pack node first and a general-purpose helper second. If you're not using the Gemini nodes, you don't need it.