参数覆盖
16.width=512 — set remote workflow params without touching the JSON
- 串联输入
- 串联输出
参数覆盖, "param override," is the non-file half of RemoteCall's injection story. Where the Transfer node ships files to a remote workflow, this node sets scalar values - width, height, seed, model names - using a one-line-per-value text format that's about as compact as it gets:
16.width=512
16.height=512
10.model_name=RealESRGAN_x4plus.pth
That's the whole trick: 节点ID.字段=值, one per line (or semicolon-separated if you're feeling cramped). You type those lines, the node parses them, and they end up merged into the exact inputs of the exact nodes in the remote workflow.
How it works
It takes two inputs: 参数覆盖 (the multiline text) and 串联输入 (the transfer chain from a Transfer node, or from another override further up). It walks your lines, infers a type for each value - anything with a . or an e becomes a float, plain numbers become ints, everything else stays a string - and merges them into the chain dict.
The clever part is what happens downstream. When 上传提交 walks the chain and finds a dict value for a node ID (which is what this node produces), it merges those keys straight into that node's inputs in the prompt JSON. No upload, no filename swapping - the value is just written in place. That's how 16.width=512 becomes the actual width of node 16 on the remote instance.
Chain it between your Transfer nodes and the 上传提交 node, and both kinds of data travel on the same single wire.
The inputs that matter
- 参数覆盖 - the only thing you really set. One
节点ID.字段=值per line. - 串联输入 - required; the chain has to come from somewhere.
Output is 串联输出, the updated chain, which continues on to 上传提交 (or into another override to layer more values).
The gotchas
Two, and both are the sort that bite once.
First, type inference is aggressive. A value like 64 becomes an integer, 0.5 becomes a float. If some remote field genuinely needs a numeric-looking string, this node will convert it on you - there's no way to force a string. For the normal cases (dimensions, seeds, steps, model names) it's exactly right.
Second, you're writing field values, not moving files. 10.model_name=RealESRGAN_x4plus.pth only works if that upscaler actually exists on the target instance's models folder. This pack uploads the files you hand it through Transfer; it does not ferry model weights around. If the remote job errors with a missing model, the model is missing over there - download it on box B, not box A.
Installing it
Part of the ComfyUI-RemoteCall pack, install once for all seven nodes: ComfyUI Manager → search ComfyUI-RemoteCall (or 远程调用), restart, or:
cd ComfyUI/custom_nodes
git clone https://github.com/playboy-dongan/ComfyUI-RemoteCall
Only requests and av as dependencies, no models to download.
Why you'd reach for it
This is the node that turns a shared remote workflow into a reusable template. You parse it once, discover which fields are the knobs, and write exactly those as override lines - without ever hand-editing the workflow JSON or asking the person who owns the remote instance to change anything. It's fiddly on day one, quietly invaluable after.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| 参数覆盖 | STRING | — | |
| 串联输入 | REMOTE_CALL_TRANSFER | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| 串联输出 | REMOTE_CALL_TRANSFER | — |