GPT img API ChatGPT LLM

You probably found this pack for the image generation, but the LLM node is quietly the more useful one. GPT img API ChatGPT LLM sends a prompt to a ChatGPT model and returns the text reply as a plain STRING - which means you can use it the way ComfyUI users use any text node: chain it into a prompt encoder, have it rewrite a negative prompt, generate a caption, or summarize what the previous node decided. Anywhere you'd stick a hand-typed string, this can feed a live model instead.

The trade-off is the API key again: every call is billed to your OpenAI account. There's no local model, no GGUF to download, no VRAM - the whole exchange happens at api.openai.com. If you'd rather use your ChatGPT subscription, the OAuth twin (GPT img OAuth ChatGPT LLM) is the same node with login-based auth.

How it works

It POSTs to the Responses API (/v1/responses) with your system_prompt as a developer message and your prompt as the user message. The payload sets reasoning to your chosen reasoning_effort and caps the reply with max_output_tokens. The response comes back as JSON, and the node extracts the text while also keeping the whole raw payload for you.

The inputs that matter

• prompt - the actual request. This is the one you'll edit constantly.
• system_prompt - role instructions; "You are a helpful assistant" is the default, and it's fine to leave it for casual use.
• model - defaults to gpt-5.5, and this node also exposes gpt-5.5-pro for heavier reasoning.
• reasoning_effort - minimal through xhigh. Medium is default; crank it for hard questions, drop it when you just want speed.
• max_output_tokens - output budget, default 2048, up to 128000.
• api_key - empty means it falls back to OPENAI_API_KEY in your environment.

Two optional sockets, system_prompt_input and user_prompt_input, override the widgets when connected - handy when you're feeding this node from another node's output instead of typing.

Outputs

• text - the model's answer, a STRING you can wire into anything.
• raw_response_json - the full response object as a string, for when you need to inspect tokens or metadata.

Install

Manager may not show this pack yet (its registry version is still Pending), so clone manually:

cd ComfyUI/custom_nodes
git clone https://github.com/hub2vu/Comfyui-GPT-img-node.git GPT-img

Restart ComfyUI, find the "GPT img" category. Zero Python dependencies to install.

Common issues

The usual suspects: a missing or invalid api_key, and long generations hitting the timeout_sec default of 300 seconds. The model list here is broader than the image nodes because pro models can have different streaming support, so if a call fails oddly, check you're not picking a model the image nodes wouldn't offer. And as always with API-key nodes - don't commit the key into a workflow file that ends up in git.