Nodes/storyboard/OpenAI Adv. Config
ComfyUI Node

OpenAI Adv. Config

The Two Knobs OpenAI Chat GPT Keeps Hidden

By laubsauger·Created about a year ago·Updated about a year ago· 10
OpenAI Adv. Config
    • ADV_CONFIG
    response_format
    seed42

    The pack's OpenAI Chat GPT node keeps its face clean: prompt, system prompt, model, temperature. But OpenAI's API has more levers, and this little node exists to expose two of them - response format and seed - without bloating the chat node's interface. It's a data-plumbing node: it builds a small config object and hands it to the chat node over a single ADV_CONFIG socket.

    How it works

    Nothing clever here. You set two fields and it returns a dict:

    • response_format - text or json_object.
    • seed (INT, default 42, range 0–4294967295) - the API's deterministic-generation seed, for models that support it.

    The node packages those into {"response_format": {"type": "..."}, "seed": ...} and outputs an ADV_CONFIG object. Wire that into OpenAI Chat GPT's optional adv_config input and the chat node merges it straight into the API request. That's the whole mechanism - there's no API call, no network, no key needed. It's a config value that becomes part of someone else's request.

    Why you'd bother

    json_object is the genuinely useful one. Set it and the model is constrained to emit valid JSON, which makes the LLM output usable as structured data downstream - feed it to a parser, a prompt-router, whatever you're orchestrating. The seed gives you reproducible completions, which is nice for A/B testing prompts without the model wiggling between runs.

    The honest caveat: OpenAI's json_object mode has historically required the word "json" to appear somewhere in the prompt, or the model can refuse or wander. If you get malformed output, check that first.

    Inputs and output

    • response_format - enum: text or json_object.
    • seed - integer.
    • Output: ADV_CONFIG - wire it to the matching socket on OpenAI Chat GPT. That's its only consumer in this pack.

    Install

    It's one of five nodes in the comfyui-storyboard pack, so you install once:

    # ComfyUI Manager → search "comfyui-storyboard" → Install → Restart
    # or
    cd ComfyUI/custom_nodes
    git clone https://github.com/laubsauger/comfyui-storyboard
    

    Notes

    This is a config object, not a magic wand. If you don't need JSON output or reproducibility, skip it entirely - the chat node works perfectly fine without an adv_config hooked up. And the two fields are frozen in time; if you need newer API options than response format and seed, this node won't get you there until the pack updates.

    Categorystoryboard/LLM/OpenAI

    Inputs (2)

    NameTypeDefaultDescription
    response_formatCOMBO2 options: text, json_object
    seedINT420–4294967295

    Outputs (1)

    NameTypeDescription
    ADV_CONFIGADV_CONFIG