MultiTrack Prompt Enhancer Image List Bridge
The unglamorous bridge node your local llama.cpp prompt graph needs
- images
- IMAGES
Some nodes are stars, and some are plumbing. MultiTrack Prompt Enhancer Image List Bridge is plumbing, and the pack labels it exactly that way: _easy_media/internal. You will almost never add it to a graph by hand - it gets pulled in automatically when you build the local llama.cpp prompt-enhancement path and the visual editor needs a way to keep image lists intact through the graph.
What it actually does
When you use MultiTrack Prompt Enhancer with the local llama.cpp provider, the images that need describing are a list - multiple reference frames from a segment. The bridge node preserves that list-style input on its way into the vision model, rather than letting it collapse into a single batch. Its three inputs are the whole story:
images- the image values forwarded by MultiTrack Prompt Enhancer.max_size- the long-edge size (128–768, default 512) images are resized to before local vision inference. Smaller = faster and cheaper VRAM, bigger = more detail for the model to read.inference_mode- "one by one" (default), which keeps each reference separate and shares the visual-size budget across the images or sampled video frames.
The single output is IMAGES, a proper ComfyUI list, ready for the local VLM.
That's it. If you're using the MiniMax or third-party API paths of the enhancer, this node never appears and you don't need to think about it.
The practical reality
Because it's internal, don't build around it directly. If it shows up in your workflow, treat it as a signal: you're on the local llama.cpp path, which means you need ComfyUI-llama-cpp_vlm (the llama_cpp_instruct_adv support) installed, or the chain breaks with a missing-package error. If you're debugging a local prompt-enhancement graph and things stall at this node, the usual culprits are the max_size budget (512 is a sane default; pushing it to 768 makes vision slower without reliably better prompts) or a VLM that wasn't actually loaded.
Install
It's in ComfyUI-Easy-Media - install the pack, restart ComfyUI:
cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Media.git
There's genuinely nothing else to configure here. The node's whole existence is "keep the list shaped right," and if everything upstream is happy, it's invisible. When a tool works so quietly you forget it's there, that's the design working - this one just asks that you not delete it from the graph when you're tidying up your workflow.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| images | * | Image values forwarded by MultiTrack Prompt Enhancer. | |
| max_size | INT | 512128–768 | Maximum image long-edge size used before local llama.cpp vision inference. |
| inference_mode | COMBO | one by one | Preserves separate references and distributes the visual-size budget across images or sampled video frames. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGES | IMAGE | Images forwarded as a ComfyUI output list. |