ComfyUI_OpenAI_GPTImage2

ComfyUI OpenAI GPT-Image-2 Nodes

ComfyUI custom nodes for OpenAI GPT Image generation and image editing.

Nodes

• GPT-Image-2 Text-to-Image: calls the Images API with gpt-image-2.
• GPT-Image-2 Image Edit: calls the Images API edit endpoint with gpt-image-2.
• GPT-Image-2 Image Edit (Responses API, file cache): uploads input images as files, reuses cached file_id values in the current ComfyUI process, and calls the Responses API with the image_generation tool.

All nodes keep the same outputs:

• image: generated image tensor
• url: (base64) for base64 image results
• response: request summary or error text

Model Types

gpt-image-2 and gpt-5.x models are not the same kind of model.

| Field | Typical value | Used by | Purpose | | --- | --- | --- | --- | | image_model | gpt-image-2 | Images API or image_generation tool | Renders generated or edited images | | responses_model | gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano | Responses API | Understands the request and calls tools |

For the T2I and Images Edit nodes, the node directly uses gpt-image-2.

For the Responses Edit node, choose:

• responses_model: the main Responses API model. Default: gpt-5.5.
• image_model: the GPT Image model used by the image_generation tool. Default: gpt-image-2.
• image_model = auto: omit the tool model and let the API choose.
• custom_responses_model / custom_image_model: use these only when the dropdown is set to custom.

Install

Install the requirements into the same Python environment used by ComfyUI:

E:\ComfyUI_windows_portable\python_embeded\python.exe -m pip install -r requirements.txt

Then restart ComfyUI.

API Key

Each node keeps api_key as a node input. Paste your OpenAI API key into the node before running the workflow.

Proxy And TLS

The optional proxy field accepts values such as:

• http://127.0.0.1:7890
• socks5://127.0.0.1:1080

The client currently uses verify=False in httpx.Client to improve compatibility with local proxy setups that intercept TLS. Keep this in mind when using untrusted networks or proxies.

Size Limits

Custom output size is validated before the API request:

• Width and height must be positive.
• Width and height must be multiples of 16.
• Maximum edge length is 3840px.
• Total pixels must be between 655360 and 8294400.
• Aspect ratio must be no more than 3:1.

Examples:

• 1024x1024: valid
• 1536x1024: valid
• 2048x2048: valid
• 3840x2160: valid
• 256x256: invalid because total pixels are too low
• 3840x3840: invalid because total pixels are too high

Common Errors

• API Key Required: fill the api_key field.
• Prompt Required: enter a non-empty prompt.
• Invalid Size: adjust width and height to satisfy the size limits above.
• Too Many Images: edit nodes support up to 16 input images.
• Connection Failed: set or check the proxy if your network cannot reach the OpenAI API directly.
• No Image in Response: the API returned successfully but did not include image data; check the model fields and prompt.

Notes

• The Responses Edit node cache is in memory only. Restarting ComfyUI clears cached file_id values.
• Masking with GPT Image is prompt-guided. The mask helps guide edits, but exact mask boundaries are not guaranteed.
• seed inputs are kept for workflow compatibility and metadata, but the current API calls do not pass a seed parameter.