ComfyUI Node
NInfer - Client
Connection to a local NInfer server. Starts the server on first use and lets it shut down again once the queue has been idle.
NInfer - Client
- Client
◄base_urlhttp://127.0.0.1:8080/v1►
◄auto_starttrue►
◄model_fileHuihui-Qwen3.8-27B-abliterated-Swift-1.5-GSQ-RCO-IQ3_S-mtp.ninfer►
◄idle_seconds120►
◄keep_alivefalse►
◄api_keyninfer-local►
◄max_retries1►
◄timeout900►
CategoryNInfer
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| base_url | STRING | http://127.0.0.1:8080/v1 | NInfer's OpenAI-compatible endpoint |
| auto_start | BOOLEAN | true | Start the server if it is not running. Turn this off to use a server you launched yourself with infer.ps1. |
| model_file | COMBO | Huihui-Qwen3.8-27B-abliterated-Swift-1.5-GSQ-RCO-IQ3_S-mtp.ninfer | Artifact to load when this node starts the server |
| idle_seconds | INT | 12030–3600 | Stop the server after this long with no request. NInfer cannot unload a model, so stopping the process is the only way to give the VRAM back to ComfyUI. |
| keep_alive | BOOLEAN | false | Never shut down automatically |
| api_key | STRING | ninfer-local | NInfer itself ignores this header entirely, so the default placeholder is fine locally. It is here so the same nodes can talk to any OpenAI-compatible endpoint that does check credentials. Never leave it empty: an empty api_key makes the openai client fail to connect. |
| max_retries | INT | 10–10 | Retries on connection errors and 5xx responses |
| timeout | INT | 9001–3600 | Whole-request budget. Generations here run for minutes, so keep this generous. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| Client | NINFER_CLIENT | Pass to a NInfer chat node |