Custom ComfyUI node that controls a local Codex agent from ComfyUI.
Codex Generate Image
- Widget:
prompt - Optional inputs:
image,image_2,image_3,image_4 - Output:
IMAGEbatch - Output:
image_paths - Output:
codex_response
The node starts Codex through the experimental Python SDK, asks it to generate one image file in non-interactive mode, parses the final JSON response, loads the returned file path, and returns it as a ComfyUI IMAGE.
If image inputs are connected, the node saves them under codex_reference_inputs and passes their file paths to Codex as reference/edit source images. Batched ComfyUI image tensors are expanded into multiple reference files.
The number of output images is determined by the prompt. If the user asks for an animation, sequence, storyboard, or frames, Codex should generate multiple ordered images; otherwise it should generate one image. The node loads every returned path as one ComfyUI image batch.
The default model settings are gpt-5.5 with reasoning effort medium.
The width and height widgets support gpt-image-2 popular dimensions:
1024x10241536x10241024x15362048x20482048x11523840x21602160x3840auto
The node validates the selected pair against gpt-image-2 constraints before calling Codex.
Codex Generate Video
- Widget:
prompt - Optional inputs:
image,image_2,image_3,image_4 - Output:
VIDEO - Output:
video_path - Output:
project_dir - Output:
codex_response
The video node starts Codex through the same SDK path, asks it to create a HyperFrames project, optionally uses the image generation skill for raster assets, renders the composition, validates the returned file path, and returns it as a ComfyUI VIDEO.
The default target is a 6 second, 1920x1080, 30 FPS MP4 rendered at HyperFrames standard quality. The width, height, duration_seconds, fps, format, and quality widgets are passed to Codex as render requirements.
If image inputs are connected, the node saves them under codex_reference_inputs and passes their file paths to Codex as reference/edit source images for the video project.
The Codex Generate Video node uses HyperFrames for HTML-based video composition and rendering:
https://github.com/heygen-com/hyperframes
HyperFrames is licensed under the Apache License, Version 2.0.
This node is intended for local ComfyUI workflows where the person running ComfyUI also controls the local Codex authentication/runtime.
Do not expose this node as a public or shared service that runs against your personal Codex session. If multiple people use this node, each user should authenticate and run Codex with their own authorized account or workspace controls.
Codex can read files and run commands in the selected working directory. Use a trusted working directory, review generated files before reuse, and avoid passing private or sensitive data unless you are comfortable sharing it with the configured Codex/OpenAI service.
Prompts and generated outputs must comply with OpenAI's terms and usage policies:
Do not use this node to bypass Codex limits, safeguards, or approval flows.
Install the Codex Python SDK in the same Python environment used by ComfyUI:
python -m pip install -r requirements.txtThe requirement installs the Python SDK from the sdk/python subdirectory of the open-source Codex repo. The official docs also support installing from a local Codex checkout:
cd sdk/python
python -m pip install -e .Use the python executable from the same environment that starts ComfyUI. For example, this may be a virtualenv, Conda environment, or the embedded Python bundled with a portable ComfyUI install.
The node does not require OPENAI_API_KEY. It uses your local Codex authentication/runtime.
The node will try to find codex from PATH automatically. If ComfyUI starts with a different PATH, set the node's codex_bin widget to the absolute path of your local codex executable.