Ino Openai Responses

OpenAI's newer Responses API is the one that took over from chat completions for current models - it's the unified endpoint that returns status, reasoning, and structured output instead of just a message blob. InoOpenaiResponses is a direct wrapper around it, and its party trick is response_type: text or image. Set it to image, pass a URL in image_url alongside your text, and you're sending a vision request where the model can look at an actual image and answer questions about it - all inside the graph, with the answer coming back as a string you can route anywhere.

It's the sibling of Ino Openai Chat Completions (which is the "any OpenAI-compatible endpoint, including vLLM/RunPod" one). This node is more opinionated: it's built for OpenAI proper and its Responses API, with a model dropdown limited to gpt-5 and gpt-4.1. If you're already paying for OpenAI, it's the cleaner fit.

How it works

Required inputs: enabled, seed (an INT with a "control after generate" toggle, so you can randomize or hold it), response_type (text or image), text (your prompt), and image_url (used when response_type is image).

Optional: openai_api_key (falls back to the OPENAI_TOKEN env var if empty), timeout (default 300s), max_retries (default 3), and model.

Outputs: success, id (the response ID), status, error, output_text (the model's answer as plain text - the one you'll wire into things), and output (the full serialized response structure, for when you need the details). In image mode, the request becomes a user message with both input_text and input_image content parts, which is how the model actually sees the picture.

Installing it

Ships with ComfyUI-InoNodes by nobandegani:

cd ComfyUI/custom_nodes
git clone https://github.com/nobandegani/comfyui_ino_nodes
cd comfyui_ino_nodes
pip install -r requirements.txt

Or search "ComfyUI Ino Nodes" in ComfyUI Manager, install, restart. Needs the pack's openai dependency and a current ComfyUI (V3 schema).

Common issues

The usual suspects: no key set (paste it in or set OPENAI_TOKEN - the node reads the env var first only if the field is empty), and image URLs that the API can't fetch. image_url must be a publicly reachable URL or a data URI the model can actually load - a file:// path or a localhost address will fail. Also remember output_text is the field to feed your graph; output is the verbose structure and will include reasoning parts you probably don't want going into a prompt. And model being a dropdown means you're stuck with what's listed - if you need a different model, the chat-completions sibling lets you type any string.