Nodes/ERPK Collection/OpenAI Image Generation (Responses)
ComfyUI Node

OpenAI Image Generation (Responses)

A ComfyUI node in ERPK/OpenAI with 13 inputs and 3 outputs.

By eRepublik-Labs·Created 10 months ago·Updated about a month ago· 0
OpenAI Image Generation (Responses)
  • client
  • image
  • revised_prompt
  • reasoning_summary
prompt
seed-1
mainline_modelgpt-5.5
image_modelgpt-image-2
reasoning_effortnone
verbositydefault
size1024x1024
qualityauto
backgroundauto
output_formatpng
moderationauto
enable_web_searchfalse
CategoryERPK/OpenAI

Inputs (13)

NameTypeDefaultDescription
promptSTRINGImage description. The mainline model may auto-revise it before passing to the image model.
seedINT-1-1–2147483647Cache-bust seed. Randomizes by default.
clientoptOPENAI_API_CLIENTOpenAI API client from OpenAI API Config node
mainline_modeloptCOMBOgpt-5.5Text/reasoning model that drives the Responses API call. Not the image model — this picks prompt interpretation, reasoning, and (optionally) web search. gpt-5.5 is the current premium flagship with the highest reasoning tier; gpt-5.4 is a cheaper alternative for cost-sensitive workflows.
image_modeloptCOMBOgpt-image-2Underlying GPT Image model used for pixel generation inside the tool.
reasoning_effortoptCOMBOnoneMainline-model reasoning depth. 'none' skips reasoning (cheapest, fastest). 'low' to 'xhigh' increase prompt-interpretation quality at token cost. Only supported by reasoning-capable mainline models (gpt-5.x, o3, o4-mini). Note: when enabled, reasoning_summary output contains the model's raw chain-of-thought, which often includes orchestration-level thinking (output channels, response format) rather than only creative rationale.
verbosityoptCOMBOdefaultMainline-model output verbosity (gpt-5.x family). Shapes how chatty the response is independent of max_tokens. 'default' lets the model pick. Silently dropped for older mainlines.
sizeoptCOMBO1024x1024Image size. Same constraints as direct endpoint (gpt-image-2: min 655,360 pixels).
qualityoptCOMBOautoImage quality tier.
backgroundoptCOMBOautoBackground type. gpt-image-2 rejects 'transparent' — auto-coerced to 'opaque' with a warning log.
output_formatoptCOMBOpngOutput image format.
moderationoptCOMBOautoContent moderation level. 'low' relaxes default safety filters.
enable_web_searchoptBOOLEANfalseAdd the web_search tool alongside image_generation. The mainline model can decide to look up reference material before generating. Adds $10/1000 calls when the model actually invokes it.

Outputs (3)

NameTypeDescription
imageIMAGE
revised_promptSTRING
reasoning_summarySTRING