EmAySee Submit to Oobabooga API Node With API Key

This is the sibling of EmAySee Submit to Oobabooga API with exactly one behavioral difference: it adds an apikey_input and sends it as an Authorization: Bearer <key> header. That's it. If your text-generation-webui is configured to require an API key (openai-style auth), the no-key version returns 401s and this one works. If your server doesn't need a key, this version works too - most servers ignore a Bearer token they don't validate, though you might as well use the no-key node then.

Everything else is inherited from its sibling: same completions endpoint shape, same sampler knobs, same stream: false, same error handling that returns error strings instead of crashing the graph. The real message of this node is "the pack author's servers are behind a key," and by extension, this is the version you want once your own local LLM setup has any sort of auth on it.

The inputs that matter

• apikey_input - the token your text-generation-webui expects. Default is the placeholder text "Enter your api key here", so the first run will fail if you forget this.
• api_endpoint - change this. The default is http://10.0.0.57:5000/v1/completions, a private LAN address from the author's own setup. Set it to http://127.0.0.1:5000/v1/completions or wherever your server actually lives.
• text_input - the prompt sent to the LLM.
• max_new_tokens - dropdown, default 300 (this variant's list runs up to 4096).
• temperature, top_p, top_k, typical_p, repetition_penalty, seed, delay_seconds, show_api_response - same story as the no-key version.

Output is a single api_response string, wired into your CLIP Text Encode or a text combiner. It's an output node, so it runs as a terminal step.

Installing it

Same pack as always: ComfyUI_EmAySee_CustomNodes via ComfyUI Manager, or git clone https://github.com/EmAySee/ComfyUI_EmAySee_CustomNodes into custom_nodes/, restart. No extra Python deps, no model downloads.

Where people get burned

Aside from the LAN-IP default (change it, really), the usual suspects: your text-generation-webui must have its API enabled and the key configured in its settings - a wrong key gets you a 401 that surfaces as an error string in the output. And remember this is the completions endpoint (/v1/completions), not /v1/chat/completions - if you've been playing with the chat-flavored nodes in this pack, the path is easy to mix up, and a wrong path is the most common 404 you'll see. If you're doing chat-style "system prompt + user message" calls with reasoning models, skip this node and use the _WithThinkParse variant instead - it speaks chat/completions and splits thinking traces out of the reply.