VividAI 配置

VividAIConfig (显示 "VividAI 配置") is the smallest node in this pack and the one you cannot skip. It's a credential holder that plugs into the three generation nodes. Your API key lives in a local SQLite file on your machine, the node carries only a reference to that profile, and the workflow you upload to the internet contains no secret at all.

Why a whole node for a key

Everything VividAI does is an HTTP call. There is no model on your disk, no checkpoint, no VRAM curve - you send a prompt to vividai.run, it generates on its hardware and hands back pixels. This is the API-wrapper corner of ComfyUI, the same shape as the Partner Nodes and every ComfyUI-<vendor>-API pack, and it comes with the whole category's baggage: per-call cost, your inputs leaving the machine, and a node that is supposed to phone home by design.

The last point is why this node exists in this form. An API node holds a credential and makes outbound calls, which is exactly the pattern that has been weaponized in this ecosystem before. So the author split the two concerns: the network node never touches the key, and the key never enters a widget value.

There's also zero community footprint for this pack as of writing - no Reddit threads for "VividAI" or "Neko1t" in the corpus, no wiki entry, nothing. Judge it on the code, not on reputation - and note below that the local database it writes is a secret you can't put in a backup you share.

How it works

The node has one required input, config_ref - the author's own tooltip is "本机配置引用，不保存 API Key" (a reference to a local profile; it does not store the API key). In the UI it doesn't look like a text field: the pack's frontend extension replaces it with a dropdown of your saved local profiles, labelled 本机配置. Under the hood execute() just looks the profile up in the local store and passes the reference through.

One output: 配置 (VIVIDAI_CONFIG), which goes into the config input on VividAI 文生图, VividAI 文生视频 or VividAI 参考素材生视频.

A detail worth knowing: that config input is declared lazy, and the generation nodes only request it when they're actually going to submit a new job. If a node is reusing a cached material, the config branch isn't evaluated at all. Nothing here is expensive, but it tells you the design intent - the credential is a local lookup, not workflow data.

Your profiles live in a small database (state.sqlite3) alongside the plugin's cache, in ComfyUI's user directory for vividai (the README calls it ComfyUI/user/__vividai). The file is created with 0600 permissions, and the API only ever returns profiles with the secret stripped out.

Configuring it the first time

1. Add the node to the canvas and click 管理本机 API 配置.
2. Give the profile a name, set the base URL to https://vividai.run, and paste the key into the separate password box (that's intentional - the key is never a widget value, so it can't get serialized into the graph).
3. Save, then pick the profile in the 本机配置 dropdown.

The base URL is validated as an HTTPS root - host and port, no path, no query, no embedded credentials - and plain http:// is accepted only for loopback, so you can point it at a local test proxy but not at something random on your LAN.

Install

In ComfyUI Manager, search ComfyUI VividAI - that's the pack's registered display name, and it's the thing to search for, because the GitHub repo is confusingly named ComfyUI-plugin. Or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/Neko1t/ComfyUI-plugin
cd ComfyUI-plugin

Then install the dependencies into the Python ComfyUI actually runs, not some other interpreter:

uv pip install --python /path/to/ComfyUI/python -r requirements.txt

Windows portable builds usually mean python_embeded\python.exe. The four requirements are real and all used: httpx, Pillow, numpy, and av (PyAV, which does the video result validation). No PyTorch - it uses your host's copy. Restart ComfyUI afterwards.

Per the README, put the whole repo in custom_nodes, not just the inner vividai package folder. That mistake is the most common way this ends in an import error.

Troubleshooting

The management panel refuses to open (403). The credential routes only accept loopback connections from the same origin. If you reach ComfyUI through a LAN IP, a reverse proxy, a tunnel, or a container with a non-localhost host header, 管理本机 API 配置 will 403 by design. Open http://127.0.0.1:8188 directly to save the key.

"本机未找到此连接配置" (profile not found). The config_ref points at a profile that isn't in this install's database. Happens when you move a workflow between machines - the graph moves, the local database doesn't. Re-create the profile and re-pick it.

Share the workflow, not the data directory. The database contains the API key. Backing up the whole vividai folder is fine for yourself, but it must never go into a shared zip, a git repo, or a bug report. If you want the data somewhere else, set VIVIDAI_DATA_DIR.

What you're betting on. The pack is MIT, version 0.1.5, by a single author (publisher id neko1t). The README is unusually candid about verification: real workflows and the browser UI were tested on Windows against a mock service, and there has been no real paid VividAI generation run, no Linux/macOS validation, and no GPU validation. It also targets ComfyUI v0.41.3 with frontend 1.55.17 as its baseline. That honesty is a good sign - but it does mean you're an early user, so start with a key that has a spending cap on it.