Modelverse Input Files

Most of the UCloud Modelverse pack is about generating pixels, but this node is the odd one out: it feeds documents to a chat model. Modelverse Input Files picks a file out of your ComfyUI input folder and packages it so the pack's Modelverse Chat node can hand it to a cloud LLM as context. Drop a manual, a spec, a chunk of research notes next to your graph, and the model reads it before answering.

It's a small utility, but it's the difference between asking a model a question in a vacuum and asking it to work on your actual material. Think of it as the ComfyUI-shaped version of "upload a file to ChatGPT" - except it lives in the graph, so you can wire it into an automated pipeline and re-run it with new files.

What it does

The node scans your ComfyUI input directory and lists every .txt, .md, and .pdf file under 32 MB. You pick one from the file dropdown. It reads the file, base64-encodes it as a data: URI, and emits an OPENAI_INPUT_FILES output.

The genuinely clever bit is the chaining: there's an optional OPENAI_INPUT_FILES input on the node itself. Plug one instance's output into the next instance's input, and files accumulate - the tooltip spells it out: "Allows chaining of input files so that a single message can include multiple input files." So a workflow can bundle a spec PDF plus a markdown brief plus a notes file into one context.

That output feeds the files input on the Modelverse Chat node, which prepends each file's contents to the user message before calling the API. The model then answers with all of it in context.

What to watch

The dropdown only shows files already sitting in the input folder. Drop a file in there (or via the browser upload widget) and refresh, and it appears. The node checks the extension and size at listing time, so a 40 MB PDF simply won't show up - keep files under 32 MB.

A couple of grounded quirks worth knowing:

• Text and markdown get read as plain text. PDFs are accepted by the file picker; whether the upstream model extracts them well depends on the cloud model, not on this node.
• It's built for the Modelverse Chat node's OpenAI-compatible endpoint, so don't expect it to wire into arbitrary local LLM nodes.
• Files are sent up to a cloud API. If you're feeding proprietary docs, that's a data-leave-your-machine decision - the same tradeoff as any hosted chat.

Install and first run

The pack installs once and gives you every node in it: ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or git clone https://github.com/ucloud/comfyui-plugin into ComfyUI/custom_nodes. No models to download. You do need a Modelverse API key wired in via the Modelverse Client or Secret Client node, and note that sending documents is metered like everything else - a long PDF costs more tokens than a one-line prompt, so it's worth being deliberate about what you attach.