GPT Image 2 is now here via Partner Nodes
OpenAI's reasoning-powered image model — with 2K output, flawless text, and 8-image consistency — lands in ComfyUI as a Partner Node.
OpenAI’s newest image model, GPT Image 2.0 (gpt-image-2), is now available in ComfyUI through Partner Nodes. Announced today, it’s the first image model from OpenAI that reasons before it generates.
Reasoning-Driven Generation
GPT Image 2.0 is built around a thinking step. Instead of one-shot sampling, the model plans the composition, checks its work, and iterates — which is why it handles the things that have historically broken image models. Dense text, small UI elements, iconography, infographics, maps, slides, even manga panels render cleanly. For ComfyUI users, this means you can finally ask for “a poster with a seven-item bulleted list in 11pt Helvetica, centered” and get usable output instead of glyph soup.
Image Editing That Preserves What Matters
Edit-based workflows have been a weak spot across every major image model — small changes tend to ripple outward, warping faces, altering composition, or quietly rewriting details the user never asked to touch. GPT Image 2.0 handles targeted edits with structural fidelity, keeping everything outside the edit zone pixel-stable while applying the requested change cleanly at up to 2K resolution. Colorizing a black-and-white photo or shifting a scene from noon to dusk both land cleanly, without the usual collateral damage to faces, geometry, or fine detail. For retouching workflows, archival restoration, or iterative creative direction, this is a meaningful unlock.
Eight Consistent Images, One Prompt
The model can return up to eight distinct images from a single prompt while preserving character and object continuity across the series. Storyboarding, reference sheets, character turnarounds, and product variants that used to require careful seed-locking and prompt gymnastics now come out of one node. Feed the batch straight into a Save Image loop or chain it into video workflows downstream.
Getting Started
Update ComfyUI to the latest version v0.19.4, or visit Comfy Cloud
Find the node — search for OpenAI GPT Image 1.5 in the Node Library. Select the
gpt-image-2model.
GPT Image 2.0 slots naturally into hybrid pipelines: use it for the text-heavy hero frame, then hand off to your local models for upscaling, stylization, or video. That’s the point of Partner Nodes — the best model for each step, in one graph.
Let us know what you build.
I can't seem to get 'Background > Transparent' to work. Is that a feature that isn't available for Comfy Cloud users? Will it ever be implemented for the Cloud?